Research ArticleDevelopment of a Wearable Device for Motion CapturingBased on Magnetic and Inertial Measurement Units

Bin Fang Fuchun Sun Huaping Liu and Di Guo

State Key Lab of Intelligent Technology and Systems Tsinghua National Laboratory for Information Science andTechnology Department of Computer Science and Technology Tsinghua University Haidian District Beijing 100083 China

Correspondence should be addressed to Bin Fang fangbin1120163com

Received 22 July 2016 Revised 26 September 2016 Accepted 17 October 2016 Published 18 January 2017

Academic Editor Michele Risi

Copyright copy 2017 Bin Fang et alThis is an open access article distributed under the Creative Commons Attribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

This paper presents a novel wearable device for gesture capturing based on inertial and magnetic measurement units that are madeup of micromachined gyroscopes accelerometers and magnetometers The low-cost inertial and magnetic measurement unit iscompact and small enough to wear and there are altogether thirty-six units integrated in the deviceThe device is composed of twosymmetric parts and either the right part or the left one contains eighteen units covering all the segments of the arm palm andfingers The offline calibration and online calibration are proposed to improve the accuracy of sensors Multiple quaternion-basedextended Kalman filters are designed to estimate the absolute orientations and kinematic models of the arm-hand are consideredto determine the relative orientations Furthermore position algorithm is deduced to compute the positions of corresponding jointFinally several experiments are implemented to verify the effectiveness of the proposed wearable device

1 Introduction

The gesture is a natural and efficient way for communicationand plays an important role in human-human interaction toexpress and transmit information Therefore the technologyof gesture recognition has become a hot research topicThrough gesture recognition results are used to commu-nicate with computers assess the kinematics of the handcontrol electronic equipment and so on It has been wildlyused in several application areas such as rehabilitationsports and animation industry [1 2] In order to identifyhuman gestures we need to obtain the positions speedsdirections and other information of the gesture by a certaingesture capture technology

At present there are two main kinds of motion capturetechnology namely vision based and contact based devices[3 4] Vision based devices capture the video streams foranalysis to determine the hand motion On the other handcontact based devices depend on physical interaction withthe user Based on vision gesture method users generallydo not need to wear collection equipment and can movemore freely But it is more susceptible to backgroundlike illumination occlusion and other environmental fac-tors Furthermore the camera frame rate and placement

requirements are higher [5] In comparison contact baseddevices are easy to implement Examples of contact baseddevices are mobile touch screens EMG-based devices anddata gloves EMG-based device is portable but EMG signalis easily influenced by acquisition placement individualdifferences physical condition and other factors Moreoverfine finger and hand motion is still difficult to determine [6]Data gloves use multiple sensor such as optical fiber sensorpressure resistance sensor pressure sensor magnetometeraccelerometer and gyroscope to perceive the movement [7]This kind of method can well reflect the spatial motiontrajectory attitude and time sequence information of thehand by capturing the position direction and angle of thefinger which are restricted to the environmental conditionsBut at present the product price is very expensive so thedevelopment of low-cost data gloves has become the goalof our research Different types of sensory gloves have beendeveloped overtime both commercial and prototype onesThe commercial products [8] usually use expensive motion-sensing fibers and resistive-bend sensors and are too costlyfor the consumer market [9] Consequently the prototypedata gloves are developed to lower the cost of such equipment[10] The flex sensors or bending sensors are integrated intothe data gloves However the above sensors just measure the

HindawiScientific ProgrammingVolume 2017 Article ID 7594763 11 pageshttpsdoiorg10115520177594763

2 Scientific Programming

relative orientation of articulated segments by mounting thesensor across the joint of interest This requires an accuratealignment of sensors with a particular joint Moreoverrecalibration is necessary to mitigate estimation errors dueto the sensor displacements General disadvantages of thedata gloves are the lack of user customization for individualsubjectrsquo hands and obstruction of tactile sensing from thepalmar surface of the hand Often this inherently goes withthe lack of mounting space for embedding the sensors inthe cloth To overcome these shortcomings the inertial andmagnetic sensors are induced

In recent years the MEMS technology has developedtremendously The microinertial sensors have so manyadvantages like low-cost small size low-power consumptionlarge dynamic range and so on It has gradually become oneof the most popular sensors for human motion capturing[11] Meanwhile the magnetic sensors are commonly usedtogether with inertial sensors for accurate and drift-free ori-entation estimation [12] Inertial and magnetic measurementunit (IMMU) has been proved to be an accurate approachto estimating the orientations of body segments without theexternal cameras [13] It is nonobtrusive comparably costeffective and easy to setup and use It also demonstrateshigher correlation and lower error compared with a research-used visual motion capture system when the same motionsare recorded [14] Furthermore the wearable inertial andmagnetic sensors are becoming increasingly popular for thegesture motion capturing

Gesture motion capture device based on microinertialsensors can generally be divided into three types whichare hand-hold type [15] wrist-wearable type [16] and glovetype [17] Compared with the hand-hold and wrist-wearabletypes the glove type is available for having more numbersand types of inertial sensors and provides more accurateresults of gestures Therefore the forms of gloves are mostcommonly used The KHU-l data glove [17] consists of sixthree-axis accelerometers but it can only capture severalkinds of gestures In [18] the data glove is developed basedon sixteen microinertia sensors which can capture themovements of each finger and palm but the informationof the heading angle is missing In [19] the inertial andmagnetic measurement unit is used but it only uses fourinertial and magnetic measurement units which is unableto obtain the information of each finger joint Power Glove[20] is developed which includes six nine-axis microinertialsensors and ten six-axis microinertial sensors It covers eachjoint of the palm and fingers and motion characteristics canbe better evaluated However it does notmake full use of ninemicroinertial sensors and in some state the heading anglesolution is instable so that it may lead to the estimation errorsof the joint angleThe research shows that the current gesturecapture device does not take into account the motion of thearm and at the same time the movement of the hand cannotbe fully capturedTherefore we use the inertial and magneticmeasurement units to develop a new gesture capture devicewhich can fully capture themotion information of the fingershands and arms

On the other hand the IMMU-based device should befocused on the following two main aspects calibration and

fusion algorithm Calibration is important for improving theperformance of IMMU Generally there are two phases thatincluded inertial sensors calibration [21] and magnetometerscalibration in field [22] The inertial sensors are of low costand low precision the deterministic errors such as bias andscale factor exist inevitably and the calibration methodsshould be designed to improve the accuracy Moreoverthe magnetometers in the field would be affected by theiron-based materials which generate their own magneticfield Therefore the measurements of magnetometers are thecombination of earthrsquos magnetic field and the extra magneticfield caused by the environmental effects Hence magne-tometers calibration should be implemented for lesseningthe disturbance and improving the azimuth angle Fusionalgorithm is another key procedure to estimate orientationsby combining the signals of gyroscopes accelerometersand magnetometers The Kalman filter is a useful tool forsensor fusion The extended Kalman filter is [23] a generalmethod for estimating orientations and has been appliedin the products of AHRS [24] Nevertheless the EKF isnot easy to choose the appropriate parameters and hard forcomputation Then the complementary filter that combinestwo independent noisy measurements of the same signalis proposed [25] Complementary filters are the high-passsignals provided by gyroscopes and the data from low-passaccelerometers and magnetometers which provide relativelyaccurate measurements at low frequencies fused to estimatethe true orientation Reference [26] proposed a nonlinearinvariant observer with respect to the symmetries of thesystem equations In addition the deterministic algorithm isanother class for estimating the orientations

The novel wearable device proposed is developed forgesture capturing based on the IMMUs The paper is orga-nized as follows Section 2 presents the designs of thewearable device Section 3 describes the calibration of thesensors and the algorithms of attitude estimation and theposition estimation is deduced Section 4 reports the resultsof the calibration the orientations and gesture capturingexperiments which verify the effectiveness of the proposeddevice Section 5 gives the conclusions of the paper

2 Wearable Device Design

21 Inertial and Magnetic Measurement Unit Design One ofthe major designs of the wearable device is the developmentof the low-cost inertial and magnetic measurement unitsCommercial IMMU commonly contains processing unitsand transceiver modules except the MEMS inertial andmagnetic sensors This increases the weight and packagingsize hence it is not suitable to use these IMMUs to puton the right positions of the fingers or arms Moreoverit is difficult to add more IMMUs with small distance inorder to benefit from redundant measurements or gain moreaccurate measurements of fingers because of the structureand size of the unit Here the MPU9250 [27] that deployssystem in package technology and combines 9-axis inertialand magnetic sensors in a very small package is used Hencethe low-cost low-power and light-weight IMMU can bedesigned and developed It also enables powering of multiple

Scientific Programming 3

MCU

BluetoothBattery

LeftRight

N

yf23

l

xf23

l zf23

l

Y

Z

Xzr1

xr1

yr1

zl1

yl1

xl1

Figure 1 The wearable device

IMMUs bymicrocontrol unit (MCU) which reduces the totalweight of the systemMoreover small IMMU can be fastenedto the glove which makes it more convenient and easier touse The MPU9250 sensor is mounted on a solid PCB withdimensions of 10 times 15 times 26mm and a weight of about 6 g

Connectivity is another important issue in the designDifferent types of connections between sensor units andaccess points on the body have been described in [28]Wireless networking approaches for the connections are con-venient but the complexity of wireless networks is increasedand the system has to trade off energy consumption for datarates To avoid this problem a wired approach is used in [29]All sensor units are directly connected to a central controllingunit by cables which lead to a very complex wiring In thepresent work a cascaded wiring approach [30] is used anddeveloped by exploiting the master SPI bus of each IMMUThis approach simplifies wiring without any need for extracomponents Since measurements reading from a string ofIMMUs the MCU need not switch to all the IMMUs tofetch the data which leads to lower power consumptionMeanwhile the textile cables are used to connect the IMMUto each other and to the MCU for increasing the flexibilityHere the STM32F4 microcontroller is used to develop theMCU

22 Device Design After determining the above designsthe wearable design of device can be determined There arethirty-six IMMUs in the device and a pair of device isseparately put on the right hand and left hand Each side haseighteen IMMUs which covers all the segments of the armpalm and fingers Each string deploys three IMMUs and sixstrings are used Five of them are used to capture the motionsof the five fingers and the other one is used to capture themotions of the palm upper arm and forearm The batteryand MCU are attached to the wrist The wearable device isshown in Figure 1

The proposed wearable device is designed based on thelow-cost IMMUs which can capture more information of themotion than traditional sensorsThe traditional sensors usedin the data glove such as fiber or hall-effect sensors are frailNevertheless the board of inertial and magnetic sensor isan independent unit It is more compact more durable andmore robust Commercial data gloves are too costly for theconsumer market but the proposed data glove in the paper islow-cost (US$ 200) Moreover the proposed wearable device

can not only capture the motion of hand but also capturethe motion of arm and the estimated results of motion areoutputting real time

3 Methods

In this section the gesture capture algorithms are presentedFirst models of the sensors are described and the calibra-tion method is presented to improve the measurements ofthe IMMUs Then the absolute orientations filter based onquaternion-based extended Kalman filter is deduced and therelative orientations algorithm integrated kinematics of arm-hand are proposed Finally the position estimation algorithmis deduced

31 Models of the Sensors Before analyzing the models ofinertial andmagnetic sensors two coordinate frames that arethe navigation coordinate frame119873 and the body frame 119887needto be set up The orientations of a rigid body in the space aredetermined when the axis orientation of a coordinate frameattached to the body frame with respect to the navigationframe is specified According to the frames the sensorsrsquomodels are separately built as follows

(1) Rate Gyros Because the MEMS rate gyros do not haveenough sensitivity to measure the earth angular velocity themodel can get rid of the earth angular vector And the outputsignal of a rate gyro is influenced by noise and bias that is

120596119898 = 120596 + 119887119892 + 119908119892 (1)

where 120596119898 is measured by the rate gyros 120596 is the true value119887119892 is the gyrorsquo bias and 119908119892 is the noise that supposed to beGaussian with zero-means

(2) Accelerometers The measurements of accelerometers inthe body frame 119887 can be written as

a119898 = C119887119899M119886 (a + g) + b119886 + w119886 (2)

where a119898 is 3 times 3 matrixes measured by the accelerometersC119887119899 denotes the Orientation Cosine Matrix representing therotation from the navigation frame to the body frame g anda isin R3 are the gravity vector and the inertial acceleration ofthe body respectively expressed in the navigation frame and119892 = 981ms2 denotes the gravitational constantM119886 is 3 times 3matrixes that scale the accelerometers outputsb119886 is the vectorof accelerometersrsquo biasw119886 isin R3 is the vector of Gaussian withzero-means

Commonly the absolute acceleration of the rigid body inthe navigation frame is supposed to be weak 119886 ≪ 119892 or therigid body is static Then the model of accelerometers can besimplified as

a119898 = C119887119899M119886g + b119886 + w119886 (3)

(3) Magnetometers The ideal magnetic vector expressed inthe navigation frame is modeled by the unit vector Hℎ Themeasurements in the body frame 119887 are given by

h119898 = C119887119899MℎHℎ + bℎ + wℎ (4)

4 Scientific Programming

where Mℎ is a 3 times 3 matrix that scales the magnetometersoutputs bℎ denotes the disturbance vector including magne-tometersrsquo bias and magnetic effects andwℎ isin R3 is the noisesthat supposed to be Gaussian with zero-means

According to the above presentations we can know thatthe models of accelerometers and magnetometers includeerrors which would lead to errors in estimating orientationsHence the calibration should be implemented to improve theaccuracy of the sensors

32 Calibration In order to improve the accuracy of theIMMU calibration is a necessary procedure The typicalcalibration method is to assign the inertial sensors to aknown angular velocity and linear acceleration This methodnormally needs some specific equipment such as turntableHere a novel calibration method is proposed that do notneed the specific equipment and is easy to implement Thecalibration procedure has two steps First the accelerometersand magnetometers are calibrated by the offline procedureThen the gyro is calibrated by the online procedure

The unifiedmathematical models of the calibration of theaccelerometers and magnetometers can be used as follows

h119887119898 = Mℎh119887 + bℎ + w119898a119887119898 = M119886a119887 + b119886 + w119886 (5)

where triaxial sensor model is written in the vector form andbℎ b119886 are constant offset that shift the outputs of sensors

The calibration is the process that determines coefficientsbℎ b119886 Mℎ M119886 to improve the measurements of sensorsWhen the unit has omnidirectional rotation the magnitudesof the true magnetic field and gravity field remain constantand the loci of the true magnetic field measured h119887 and a119887 arespherical Meanwhile the measured h119887119898 and a

119887119898 are ellipsoids

and they can be expressed as follows [31]10038171003817100381710038171003817h119887100381710038171003817100381710038172 = (h119887119898)119879Aℎh119887119898 minus 2 (bℎ)119879Aℎh119887119898 + (bℎ)119879Aℎbℎ

+ w11989810038171003817100381710038171003817a119887100381710038171003817100381710038172 = (a119887119898)119879A119886a119887119898 minus 2 (b119886)119879A119886a119887119898 + (b119886)119879A119886b119886

+ w119886(6)

where Aℎ = (Gℎ)119879Gℎ Gℎ = (Mℎ)minus1 A119886 = (G119886)119879G119886 G119886 =(M119886)minus1 and w119898 and w119886 are assumed noiseEquations (6) are the expressions of the ellipsoid In

other words the measurements are constrained to lie on anellipsoid Thus the calibration of the magnetometers andaccelerometers is to seek ellipsoid-fittingmethods to solve thecoefficients ofG119886Gℎ b119886 bℎ And the least squares algorithmis commonly used to determine the parameters The deviceis rotated adequately to ensure that each unit gets enoughmeasurements to determine the calibration parameters byleast squares algorithm Hence the accelerometers and mag-netometers of all units of the device are entirely calibrated

After calibration the magnetometers are calibrated tolessen the environmental magnetic effect and the biases

of the sensor outputs are compensated Accordingly theprevious equation can be rewritten as follows

h119898 = C119887119899Hℎ + wℎ (7)

Since the scale and bias errors from the accelerometersare calibrated the models of the accelerometers rewritten asfollows

a119898 = C119887119899g + w119886 (8)

The offline calibration is implemented to determine thebias and scale of the accelerometers andmagnetometers Andthe online calibration is implemented to remove the gyro biasWe keep the data glove stationary for a while before use Sothe readings are zeros The bias can then be computed by themean value of the measurements The model is expressed asfollows

120596119898 = 120596 + w119892 (9)

where 120596119898 are the measurements of gyros 120596 are the trueangular velocity and w119892 are the noise of gyros

The calibration parameters of the sensors are compen-sated into the measurements so that the accuracy can beimproved for further work

33 Orientation Filters Based on these kinds of sensors twoindependent ways can determine the attitude and headingOne is determined by open-loop gyros The angular rate ofthe rigid body is measured using gyros with respect to itsbody axis frame The angles are estimated by the open-loopintegration process which has high dynamic characteristicHowever the gyro errors would cause wandering attitudeangles and the gradual instability of the integration driftingThe other way is determined from open-loop accelerome-ters and magnetometers The orientations can be correctlyobtained from accelerometers and magnetometers in theideal environment However the disturbance and noiseswould lead to large errors and make results lack reliabilityThe two ways are both quite difficult to achieve acceptableperformance Sensor fusion is a great choice to attain thestable and accurate orientations Then the fusion algorithmof quaternion-based extended Kalman filter will be deduced

The frames are shown in Figure 1 In the figure the globalframe is 119873 and local reference frame is respectively locatedin each IMMU The global reference frame 119911-axis is definedalong the axial axis (from the head to the feet) of the subject119910-axis along the sagittal (from the left shoulder to the rightshoulder) axis and 119909-axis along the coronal axis (from theback to the chest) The local frame 119911-axis is defined alongthe axial axis (normal to the surface of IMMU along thedownward) of the subject 119910-axis along the sagittal (fromthe left side to the right side of the IMMU) axis and 119909-axis along the coronal axis (from the back to the forward ofIMMU)Meanwhile two assumptions about data glove in usearemade (1) the body keeps static and only arm and hand arein motion (2) the local static magnetic field is homogeneousthroughout the whole arm

Scientific Programming 5

331 Absolute Orientation Filter Combining the measuredangular velocity acceleration and magnetic field values ofone single IMMU it can stably determine the orientationswith respect to a global coordinate system The globalcoordinate119873 and each coordinate 119887 of IMMUs are shown inFigure 1 The transformation between the representations ofa 3 times 1 column-vector x between119873 and 119887 is expressed as

x119887 = C119887119899 [q] x119899 (10)

where quaternion q = [1199020Q] [Q] is the antisymmetricmatrix given by

[Q] = [[[

0 1198763 minus1198762minus1198763 0 11987611198762 minus1198761 0]]] (11)

The attitude matrix C is related to the quaternion by

C (119902) = (11990220 minusQ sdotQ) I + 2QQ119879 + 21199020 [Q] (12)

where I is the identity matrixThe state vector is composed of the rotation quaternion

The state transition vector equation is

x119896+1 = 119902119896+1 = Φ (119879119904 120596119896) + 119908119896 = exp (Ω119896119879119904) 119902119896 + 119902119908119896119902119908119896 = minus1198791199042 Γ119896119902k119896 = minus1198791199042 [[119890119896times] + 1199024119896Iminus119890119879119896 ] 119892k119896 (13)

where the gyro measurement noise vector 119892k119896 is assumedsmall enough that a first-order approximation of the noisytransition matrix is possible

Then the process noise covariancematrix119876119896 will have thefollowing expression

119876119896 = (1198791199042 )2 Γ119896Γ119892Γ119879119896 (14)

The measurement model is constructed by stacking theaccelerometer and magnetometer measurement vectors

z119896+1 = [ a119896+1m119896+1]

= [C119887119899 (119902119896+1) 00 C119887119899 (119902119896+1)] [

gh] + [119886k119896+1119898k119896+1

] (15)

The covariance matrix of the measurement model R119896+1 is

R119896+1 = [119886R119896+1 00 119898R119896+1

] (16)

where the accelerometer and magnetometer measurementnoise 119886k119896+1 and

119898k119896+1 are uncorrelated zero-meanwhite noiseprocess and the covariancematrixes of which are 119886R119896+1 = 1205902119886Iand 119898R119896+1 = 1205902119898I respectively

Upper arm Forearm Palm Proximal Medial DistalN H

Arm Hand Finger

Y

Z X z3 z4 z5 z6x1

y1 y3 y4 y5 y6y2

z1 x2z2 x3 x6x5x4

L1 L2

Wearable position of the IMMUJoint

A1 A2 F3F1 F2L4 L5 L6L3

Figure 2 The model and frames of the arm-hand

Because of the nonlinear nature of (15) the EKF approachrequires that a first-order Taylor-Maclaurin expansion iscarried out around the current state estimation by computingthe Jacobian matrix

Η119896+1 = 120597120597x119896+1 z119896+110038161003816100381610038161003816100381610038161003816x119896+1=xminus119896+1 (17)

Then the orientations are estimated by the following EKFequations

Compute the a priori state estimate

xminus119896+1 = Φ (T119904120596119896) x119896 (18)

Compute the a priori error covariance matrix

Pminus119896+1 = Φ (T119904120596119896)P119896Φ(T119904120596119896)119879 +Q119896 (19)

Compute the Kalman gain

K119896+1 = Pminus119896+1H119879119896+1 (H119896+1Pminus119896+1H119879119896+1 + R119896+1)minus1 (20)

Compute the a posteriori state estimate

x119896+1 = xminus119896+1 + K119896+1 [z119896+1 minus 119891 (xminus119896+1)] (21)

Compute the a posteriori error covariance matrix

P119896+1 = Pminus119896+1 minus K119896+1H119896+1Pminus119896+1 (22)

According to the above algorithm the absolute orienta-tions of each IMMU can be estimated Then the kinematicsof the human arm and hand is considered

332 Relative Orientation Filter The kinematic frames ofthe arm hand and the forefinger are presented There aresix joints and the coordinate frames are built including theglobal coordinate (119873) upper-arm coordinate (1198601) forearm(1198602) palm coordinate (119867) proximal coordinate (1198651) medialcoordinate (1198652) and distal coordinate (1198653) And the lengthsbetween the joints are 1198711 1198712 1198713 1198714 1198715 and 1198716 The framesare shown in Figure 2Then the relative orientations betweentwo consecutive bodies can be determined by the following

q119903119894119895 = qminus1119894 sdot q119895 (23)

where q119903119894119895 is the quaternion of the relative orientationsq119894 is the quaternion of the absolute orientations of the

6 Scientific Programming

Gyros Accelerometers Magnetometers

Online calibration Offline calibration

Orientations estimation based QEKF

Constraint

Relative orientations

Absolute orientations

Kinematic parameters

Positions

Yes No

Kinematics of arm-hand

Figure 3 The diagram of the proposed method

first coordinate and q119895 is the quaternion of the absoluteorientations of the second coordinate

Meanwhile the kinematic models of the arm hand andfingers are considered to determine the orientations of eachsegment Human arm-hand motion can be supposed to bethe articulatedmotion of rigid body partsThese segments areupper arm (between the shoulder and elbow joints) forearm(between the elbow and wrist joints) hand (between thewrist joint and proximal joint) proximal finger (between theproximal joint and medial joint) medial finger (between themedial joint and distal joint) and distal finger (from the distaljoint) Every joint has its own local frame Shoulder can bemodeled as a ball joint with three DOFs and a fixed pointrepresenting the center of the shoulderMovements are repre-sented as the vector between the upper arm and body Elbowis the rotating hinge joint with two DOFs Wrist is modeledas a rotating hinge joint with two DOFs that is calculatedbetween the vector representing the hand and forearm Prox-imal joint is modeled as rotating hinge joint with two DOFsMedial joint and distal joint aremodeled as rotating joint withone DOF Thus the kinematic of this model consists of tenDOFs three in the shoulder joint two in the elbow joint twoin the wrist joint two in the proximal joint one in the medialjoint and one in the distal joint Hence the respondingconstraints are used to determine the orientations of eachsegment

34 Positions Estimation We assume the body keeps staticand the motion of arm and hand is formed by rotation of thejoints Hence the position of the fingertip p1198731198653 expressed in

the hand coordinate frame (see Figure 2) can be derived usingforward kinematics

[p11987311986531 ] = T1198731198601T11986011198602T1198602119867T1198671198651T11986511198652T11986521198653p11986531198652

= T1198731198653 [p119865311986521 ] (24)

where the transformation between two consecutive bodies isexpressed by T1198731198601 T11986011198602 T1198602119867 T1198671198651 T11986511198652 and T11986521198653

The total transformation T1198731198653 is given by the product ofeach consecutive contribution

T1198731198653 = [119877 (1199021198731198653) p119873119865301198793 1 ] (25)

where 119877(1199021198731198653) is the orientation of the distal phalanx withrespect to the body and p1198731198653 is the position of the distal frameexpressed in the global frame

35 Summary The gestures capture method is proposed inthe section The two procedures of the calibration are firstlyimplemented to improve the measurements of the sensorsThe inertial sensors are calibrated by the offline calibrationand the gyros are calibrated by the online calibration Thenthe orientations of the IMMUs are estimated by the QEKFAnd the kinematics of arm-hand is considered and the con-straints are integrated to determine the relative orientationsFinally the kinematic parameters are used to estimate thepositions A complete diagram of the method is depicted inFigure 3

Scientific Programming 7

minus20minus10

010

20

minus20minus10

010

20minus15

minus10

minus5

0

5

10

15z

(ms

2)

y (ms 2) x (ms2 )

Raw dataCalibrated

(a) The calibration results of accelerometers

minus1000minus500

0500

1000

minus1000minus500

0500

1000

minus1000

minus500

0

500

1000

Raw dataCalibrated

z(n

T)

y (nT) x (nT)

(b) The calibration results of magnetometers

Figure 4 The calibration results

Table 1 The calibration results

Scale Bias

Accelerometers G119886 = [[[[[

1 minus002 00 095 0120 0 092

]]]]]

b119886 = [[[[[

056087minus087

]]]]]

Magnetometers Gℎ = [[[[[

111 002 00 110 00 0 105

]]]]]

bℎ = [[[[[

571minus3954minus8298

]]]]]

4 Experiments and Results

The calibration is implemented to improve the sensors accu-racy and then the comparison experiments are testified toinvestigate the stability and accuracy of the orientations esti-mation The real-time gesture motion capture experimentsprove the validness of the proposed device

41 Calibration Results The calibration results of the triaxisaccelerometers and triaxis magnetometers are presented inthe section The calibration samples of IMMUs are collectedby rotating in various orientations Then the proposedmethod was used to determine the correctional calibrationparameters as G119886 Gℎ b119886 and bℎ The outputs of theaccelerometers and magnetometers are shown in Figure 4The blue sphere in Figure 4 is the raw date of sensors and thered sphere is the calibrated data of sensors The calibrationparameters are listed in Table 1

42 Evaluation Experiments After calibrating all the sensorsof the device the evaluation experiments are designed toassess the accuracy of the orientation estimation One of thestrings of the device is attached to the robotic arm ThreeIMMUs are set the same directions shown in Figure 5 Andthe robotic arm is rotated from the first state to the secondstate Then the estimated orientations of IMMUs compare

Table 2 The RMSE of the orientations

Static (∘) Dynamic (∘)Yaw Pitch Roll Yaw Pitch Roll

IMMU-1 004 001 002 243 005 102IMMU-2 011 001 002 168 051 103IMMU-3 010 002 002 177 002 007

with the true orientations of robotic arm The results of theorientations of three IMMUs are shown in Figure 6 Andthe results of the root mean square error (RMSE) of theorientations are listed in Table 2

As shown in Figure 6 during 5sim8 s the robotic armis dynamic And in static case it can be known that theresults have smaller variance and higher precision than theresults of the dynamic case Moreover the yaw angle has thelager variance compared with the roll and pitch It shouldbe caused by the motors of the robotic arm because themagnetometers are disturbed When the robotic arm is instatic the device can compute the accuracy orientations Thecomparison results proved the accuracy of the orientations ofthe IMMUs of the device Then the real-time gesture motioncapture experiments are implemented

43 Gesture Capture Experiments The IMMUsrsquo data is sam-pled collected and computed by the MCU and subsequentlytransmitted via Bluetooth to the external devices The MCUprocesses the raw data estimates the orientations of the eachunit encapsulates them into a packet and then sends thepacket to the PC by BluetoothThe baud rate for transmittingdata is 115200 bps The frequency is 50HZ By using thisdesign themotion capture can be demonstrated by the virtualmodel on the PC immediately The interface is written by CThe wearable system is shown as Figure 7

To further verify the effectiveness of proposed deviceupper arm is swung up and down The results of theorientations of the gestures are shown in Figure 8 And the

8 Scientific Programming

Robotic arm

IMMU-1

IMMU-2

IMMU-3

z1

x1

y1

(a) The first state

(b) The second state

Figure 5 The comparison experiments with robotic arm

positons of the fingertip of the right forefinger are shownin Figure 9 As shown in the figures the wearable devicecan determine the realistic movements of the arm andhand Meanwhile the accuracy of the results is assessed bythe statistics The root mean square error (RMSE) of theorientations is listed in Table 3 The RMSE of the positonsare listed in Table 4 Furthermore the wearable device isalso tested by ten healthy participants the real-time motiongestures capture experiments are implemented to prove thevalidity of the proposed system

44 Discussion In this paper a novel wearable device forgestures capturing based on magnetic and inertial measure-ment units is proposed As a practically useful applicationthe proposed device satisfies the requirements including theaccuracy computational efficiency and robustness First thecalibration method is designed to improve the accuracy of

0 5 10 15 20 25minus200

0

200

Time (s)

IMMU-1

YawPitch

Roll

YawPitch

Roll

YawPitch

Roll

IMMU-2

IMMU-3

Ang

el (∘

)

0 5 10 15 20 25minus200

0

200

Time (s)

Ang

el (∘

)0 5 10 15 20 25

minus200

0

200

Time (s)

Ang

el (∘

)

Figure 6 The results of orientations of the IMMUs

IMMU

Bluetooth

Data glovePC

Virtual modelTrajectory

Figure 7 The wearable system

Table 3 The RMSE of the orientations of the arm-hand

Angle (∘)Upper arm 107 110 044forearm 023 024Palm 019 018Proximal 011 026Medial 010Distal 008

Table 4 The RMSE of the positions of the fingertip

119909 (cm) 119910 (cm) 119911 (cm)Positions of the fingertip 345 135 234

the measurements of the sensors In particular the magneticeffects of the field are attenuated in advance Then the

Scientific Programming 9

0 10 20 30 40minus100

0

100Upper arm

t (s)0 10 20 30 40

minus1

0

1Forearm

t (s)

0 10 20 30 40minus1

0

1Palm

t (s)0 10 20 30 40

minus2

0

2Proximal

t (s)

0 10 20 30 40

0

05Medial

t (s)0 10 20 30 40

minus05minus05

0

05Distal

t (s)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Figure 8 The orientations of gestures

2040

6080 minus15 minus1 minus05 0 05 1

minus20

0

20

40

60

y (cm)x (cm)

z (c

m)

StartEnd

Figure 9 The positions of the fingertip of the right forefinger

orientations of gesture are divided into absolute orienta-tions and relative orientations The absolute orientations aredetermined by the QEKFs and the relative orientations areestimated by integrating the kinematics of the arm-handThe

positions are then easily computed The proposed method issimple and fast The algorithm is operated in the embeddedsystem at 50Hz The results of the experiments proved theadvantages of the proposed device

5 Conclusion

This paper presents the design implementation and exper-imental results of a wearable device for gestures captur-ing using inertial and magnetic sensor modules containingorthogonally mounted triads of accelerometers angular ratesensors and magnetometers Different from commercialmotion data gloves which usually use high-cost motion-sensing fibers to acquire hand motion data we adoptedthe low-cost inertial and magnetic sensor to reduce costMeanwhile the performance of the low-cost low-power andlight-weight IMMU is superior to some commercial IMMUFurthermore the novel device is proposed based on thirty-six IMMUs which cover the whole segments of the two armsand hands We designed the online and offline calibrationmethods to improve the accuracy of the units We deducedthe 3D arm and hand motion estimation algorithms thatintegrated the proposed kinematic models of the arm handand fingers and the attitude of gesture and positions offingertips can be determined For real-time performance and

10 Scientific Programming

convenience the interface with virtual model is designedPerformance evaluations verified that the proposed dataglove can accurately capture the motion of gestures Thesystem is developed in the way that all electronic componentscan be integrated and easy to wear This makes it moreconvenient and appealing for the user

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work has been supported by the National Science Foun-dation for Young Scientists of China (Grant no 61503212) andNational Natural Science Foundation of China (Grant nosU1613212 61327809 61210013 and 91420302)

References

[1] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) pp 25ndash28 Kitakyushu Japan August 2014

[2] J M Palacios C Sagues E Montijano and S LlorenteldquoHuman-computer interaction based on hand gestures usingRGB-D sensorsrdquo Sensors vol 13 no 9 pp 11842ndash11860 2013

[3] S S Rautaray and A Agrawal ldquoVision based hand gesturerecognition for human computer interaction a surveyrdquo Artifi-cial Intelligence Review vol 43 no 1 pp 1ndash54 2012

[4] E A Arkenbout J C F de Winter and P Breedveld ldquoRobusthandmotion tracking through data fusion of 5dt data glove andnimble VR kinect camera measurementsrdquo Sensors vol 15 no12 pp 31644ndash31671 2015

[5] D Regazzoni G De Vecchi and C Rizzi ldquoRGB cams vs RGB-D sensors low cost motion capture technologies performancesand limitationsrdquo Journal of Manufacturing Systems vol 33 no4 pp 719ndash728 2014

[6] Y Li X Chen X Zhang K Wang and Z J Wang ldquoA sign-component-based framework for Chinese sign language recog-nition using accelerometer and sEMG datardquo IEEE Transactionson Biomedical Engineering vol 59 no 10 pp 2695ndash2704 2012

[7] L Dipietro A M Sabatini and P Dario ldquoA survey of glove-based systems and their applicationsrdquo IEEE Transactions onSystems Man and Cybernetics Part C Applications and Reviewsvol 38 no 4 pp 461ndash482 2008

[8] B Takacs ldquoHow and why affordable virtual reality shapes thefuture of educationrdquoThe International Journal of Virtual Realityvol 7 no 1 pp 53ndash66 2008

[9] G Saggio ldquoA novel array of flex sensors for a goniometric gloverdquoSensors and Actuators A Physical vol 205 pp 119ndash125 2014

[10] R Gentner and J Classen ldquoDevelopment and evaluation of alow-cost sensor glove for assessment of human finger move-ments in neurophysiological settingsrdquo Journal of NeuroscienceMethods vol 178 no 1 pp 138ndash147 2009

[11] X L Wang and C L Yang ldquoConstructing gyro-free inertialmeasurement unit from dual accelerometers for gesture detec-tionrdquo Sensors amp Transducers vol 171 no 5 pp 134ndash140 2014

[12] W Chou B Fang L Ding X Ma and X Guo ldquoTwo-stepoptimal filter design for the low-cost attitude and headingreference systemsrdquo IET Science Measurement and Technologyvol 7 no 4 pp 240ndash248 2013

[13] D Roetenberg H J Luinge C T M Baten and P H VeltinkldquoCompensation of magnetic disturbances improves inertial andmagnetic sensing of human body segment orientationrdquo IEEETransactions on Neural Systems and Rehabilitation Engineeringvol 13 no 3 pp 395ndash405 2005

[14] J M Lambrecht and R F Kirsch ldquoMiniature low-powerinertial sensors promising technology for implantable motioncapture systemsrdquo IEEE Transactions on Neural Systems andRehabilitation Engineering vol 22 no 6 pp 1138ndash1147 2014

[15] R Xu S L Zhou and W J Li ldquoMEMS accelerometer basednonspecific-user hand gesture recognitionrdquo IEEE Sensors Jour-nal vol 12 no 5 pp 1166ndash1173 2012

[16] E Morganti L Angelini A Adami D Lalanne L Lorenzelliand E Mugellini ldquoA smart watch with embedded sensorsto recognize objects grasps and forearm gesturesrdquo ProcediaEngineering vol 41 pp 1169ndash1175 2012

[17] J-H Kim N DThang and T-S Kim ldquo3-D handmotion track-ing and gesture recognition using a data gloverdquo in Proceedingsof the IEEE International Symposium on Industrial Electronics(ISIE rsquo09) pp 1013ndash1018 IEEE Seoul South Korea July 2009

[18] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) vol 10 pp 25ndash28 Kitakyushu Japan August2014

[19] F Cavallo D Esposito E Rovini et al ldquoPreliminary evalu-ation of SensHand V1 in assessing motor skills performancein Parkinson diseaserdquo in Proceedings of the 2013 IEEE 13thInternational Conference onRehabilitation Robotics (ICORR rsquo13)pp 1ndash6 Seattle Wash USA June 2013

[20] H G Kortier V I Sluiter D Roetenberg and P H VeltinkldquoAssessment of hand kinematics using inertial and magneticsensorsrdquo Journal of NeuroEngineering and Rehabilitation vol 11no 1 article 70 2014

[21] Y Xu Y Wang Y Su and X Zhu ldquoResearch on the calibrationmethod of micro inertial measurement unit for engineeringapplicationrdquo Journal of Sensors vol 2016 Article ID 9108197 11pages 2016

[22] J Keighobadi ldquoFuzzy calibration of a magnetic compass forvehicular applicationsrdquoMechanical Systems and Signal Process-ing vol 25 no 6 pp 1973ndash1987 2011

[23] X Yun and E R Bachmann ldquoDesign implementation andexperimental results of a quaternion-based Kalman filter forhuman body motion trackingrdquo IEEE Transactions on Roboticsvol 22 no 6 pp 1216ndash1227 2006

[24] Xsens Technologies httpwwwxsenscom[25] T Hamel and R Mahony ldquoAttitude estimation on SO(3) based

on direct inertial measurementsrdquo in Proceedings of the IEEEInternational Conference on Robotics and Automation (ICRArsquo06) pp 2170ndash2175 Orlando Fla USA May 2006

[26] S Bonnabel PMartin and P Rouchon ldquoNon-linear symmetry-preserving observers on Lie groupsrdquo IEEE Transactions onAutomatic Control vol 54 no 7 pp 1709ndash1713 2009

[27] InvenSense MPU-9250 Nine-Axis MEMS MotionTrackingDevice 2015 httpwwwinvensensecom-productsmotion-tracking9-axismpu-9250

Scientific Programming 3

MCU

BluetoothBattery

LeftRight

N

yf23

l

xf23

l zf23

l

Y

Z

Xzr1

xr1

yr1

zl1

yl1

xl1

Figure 1 The wearable device

IMMUs bymicrocontrol unit (MCU) which reduces the totalweight of the systemMoreover small IMMU can be fastenedto the glove which makes it more convenient and easier touse The MPU9250 sensor is mounted on a solid PCB withdimensions of 10 times 15 times 26mm and a weight of about 6 g

Connectivity is another important issue in the designDifferent types of connections between sensor units andaccess points on the body have been described in [28]Wireless networking approaches for the connections are con-venient but the complexity of wireless networks is increasedand the system has to trade off energy consumption for datarates To avoid this problem a wired approach is used in [29]All sensor units are directly connected to a central controllingunit by cables which lead to a very complex wiring In thepresent work a cascaded wiring approach [30] is used anddeveloped by exploiting the master SPI bus of each IMMUThis approach simplifies wiring without any need for extracomponents Since measurements reading from a string ofIMMUs the MCU need not switch to all the IMMUs tofetch the data which leads to lower power consumptionMeanwhile the textile cables are used to connect the IMMUto each other and to the MCU for increasing the flexibilityHere the STM32F4 microcontroller is used to develop theMCU

22 Device Design After determining the above designsthe wearable design of device can be determined There arethirty-six IMMUs in the device and a pair of device isseparately put on the right hand and left hand Each side haseighteen IMMUs which covers all the segments of the armpalm and fingers Each string deploys three IMMUs and sixstrings are used Five of them are used to capture the motionsof the five fingers and the other one is used to capture themotions of the palm upper arm and forearm The batteryand MCU are attached to the wrist The wearable device isshown in Figure 1

The proposed wearable device is designed based on thelow-cost IMMUs which can capture more information of themotion than traditional sensorsThe traditional sensors usedin the data glove such as fiber or hall-effect sensors are frailNevertheless the board of inertial and magnetic sensor isan independent unit It is more compact more durable andmore robust Commercial data gloves are too costly for theconsumer market but the proposed data glove in the paper islow-cost (US$ 200) Moreover the proposed wearable device

can not only capture the motion of hand but also capturethe motion of arm and the estimated results of motion areoutputting real time

3 Methods

In this section the gesture capture algorithms are presentedFirst models of the sensors are described and the calibra-tion method is presented to improve the measurements ofthe IMMUs Then the absolute orientations filter based onquaternion-based extended Kalman filter is deduced and therelative orientations algorithm integrated kinematics of arm-hand are proposed Finally the position estimation algorithmis deduced

31 Models of the Sensors Before analyzing the models ofinertial andmagnetic sensors two coordinate frames that arethe navigation coordinate frame119873 and the body frame 119887needto be set up The orientations of a rigid body in the space aredetermined when the axis orientation of a coordinate frameattached to the body frame with respect to the navigationframe is specified According to the frames the sensorsrsquomodels are separately built as follows

(1) Rate Gyros Because the MEMS rate gyros do not haveenough sensitivity to measure the earth angular velocity themodel can get rid of the earth angular vector And the outputsignal of a rate gyro is influenced by noise and bias that is

120596119898 = 120596 + 119887119892 + 119908119892 (1)

where 120596119898 is measured by the rate gyros 120596 is the true value119887119892 is the gyrorsquo bias and 119908119892 is the noise that supposed to beGaussian with zero-means

(2) Accelerometers The measurements of accelerometers inthe body frame 119887 can be written as

a119898 = C119887119899M119886 (a + g) + b119886 + w119886 (2)

where a119898 is 3 times 3 matrixes measured by the accelerometersC119887119899 denotes the Orientation Cosine Matrix representing therotation from the navigation frame to the body frame g anda isin R3 are the gravity vector and the inertial acceleration ofthe body respectively expressed in the navigation frame and119892 = 981ms2 denotes the gravitational constantM119886 is 3 times 3matrixes that scale the accelerometers outputsb119886 is the vectorof accelerometersrsquo biasw119886 isin R3 is the vector of Gaussian withzero-means

Commonly the absolute acceleration of the rigid body inthe navigation frame is supposed to be weak 119886 ≪ 119892 or therigid body is static Then the model of accelerometers can besimplified as

a119898 = C119887119899M119886g + b119886 + w119886 (3)

(3) Magnetometers The ideal magnetic vector expressed inthe navigation frame is modeled by the unit vector Hℎ Themeasurements in the body frame 119887 are given by

h119898 = C119887119899MℎHℎ + bℎ + wℎ (4)

4 Scientific Programming

where Mℎ is a 3 times 3 matrix that scales the magnetometersoutputs bℎ denotes the disturbance vector including magne-tometersrsquo bias and magnetic effects andwℎ isin R3 is the noisesthat supposed to be Gaussian with zero-means

According to the above presentations we can know thatthe models of accelerometers and magnetometers includeerrors which would lead to errors in estimating orientationsHence the calibration should be implemented to improve theaccuracy of the sensors

32 Calibration In order to improve the accuracy of theIMMU calibration is a necessary procedure The typicalcalibration method is to assign the inertial sensors to aknown angular velocity and linear acceleration This methodnormally needs some specific equipment such as turntableHere a novel calibration method is proposed that do notneed the specific equipment and is easy to implement Thecalibration procedure has two steps First the accelerometersand magnetometers are calibrated by the offline procedureThen the gyro is calibrated by the online procedure

The unifiedmathematical models of the calibration of theaccelerometers and magnetometers can be used as follows

h119887119898 = Mℎh119887 + bℎ + w119898a119887119898 = M119886a119887 + b119886 + w119886 (5)

where triaxial sensor model is written in the vector form andbℎ b119886 are constant offset that shift the outputs of sensors

The calibration is the process that determines coefficientsbℎ b119886 Mℎ M119886 to improve the measurements of sensorsWhen the unit has omnidirectional rotation the magnitudesof the true magnetic field and gravity field remain constantand the loci of the true magnetic field measured h119887 and a119887 arespherical Meanwhile the measured h119887119898 and a

119887119898 are ellipsoids

and they can be expressed as follows [31]10038171003817100381710038171003817h119887100381710038171003817100381710038172 = (h119887119898)119879Aℎh119887119898 minus 2 (bℎ)119879Aℎh119887119898 + (bℎ)119879Aℎbℎ

+ w11989810038171003817100381710038171003817a119887100381710038171003817100381710038172 = (a119887119898)119879A119886a119887119898 minus 2 (b119886)119879A119886a119887119898 + (b119886)119879A119886b119886

+ w119886(6)

where Aℎ = (Gℎ)119879Gℎ Gℎ = (Mℎ)minus1 A119886 = (G119886)119879G119886 G119886 =(M119886)minus1 and w119898 and w119886 are assumed noiseEquations (6) are the expressions of the ellipsoid In

other words the measurements are constrained to lie on anellipsoid Thus the calibration of the magnetometers andaccelerometers is to seek ellipsoid-fittingmethods to solve thecoefficients ofG119886Gℎ b119886 bℎ And the least squares algorithmis commonly used to determine the parameters The deviceis rotated adequately to ensure that each unit gets enoughmeasurements to determine the calibration parameters byleast squares algorithm Hence the accelerometers and mag-netometers of all units of the device are entirely calibrated

After calibration the magnetometers are calibrated tolessen the environmental magnetic effect and the biases

of the sensor outputs are compensated Accordingly theprevious equation can be rewritten as follows

h119898 = C119887119899Hℎ + wℎ (7)

Since the scale and bias errors from the accelerometersare calibrated the models of the accelerometers rewritten asfollows

a119898 = C119887119899g + w119886 (8)

The offline calibration is implemented to determine thebias and scale of the accelerometers andmagnetometers Andthe online calibration is implemented to remove the gyro biasWe keep the data glove stationary for a while before use Sothe readings are zeros The bias can then be computed by themean value of the measurements The model is expressed asfollows

120596119898 = 120596 + w119892 (9)

where 120596119898 are the measurements of gyros 120596 are the trueangular velocity and w119892 are the noise of gyros

The calibration parameters of the sensors are compen-sated into the measurements so that the accuracy can beimproved for further work

33 Orientation Filters Based on these kinds of sensors twoindependent ways can determine the attitude and headingOne is determined by open-loop gyros The angular rate ofthe rigid body is measured using gyros with respect to itsbody axis frame The angles are estimated by the open-loopintegration process which has high dynamic characteristicHowever the gyro errors would cause wandering attitudeangles and the gradual instability of the integration driftingThe other way is determined from open-loop accelerome-ters and magnetometers The orientations can be correctlyobtained from accelerometers and magnetometers in theideal environment However the disturbance and noiseswould lead to large errors and make results lack reliabilityThe two ways are both quite difficult to achieve acceptableperformance Sensor fusion is a great choice to attain thestable and accurate orientations Then the fusion algorithmof quaternion-based extended Kalman filter will be deduced

The frames are shown in Figure 1 In the figure the globalframe is 119873 and local reference frame is respectively locatedin each IMMU The global reference frame 119911-axis is definedalong the axial axis (from the head to the feet) of the subject119910-axis along the sagittal (from the left shoulder to the rightshoulder) axis and 119909-axis along the coronal axis (from theback to the chest) The local frame 119911-axis is defined alongthe axial axis (normal to the surface of IMMU along thedownward) of the subject 119910-axis along the sagittal (fromthe left side to the right side of the IMMU) axis and 119909-axis along the coronal axis (from the back to the forward ofIMMU)Meanwhile two assumptions about data glove in usearemade (1) the body keeps static and only arm and hand arein motion (2) the local static magnetic field is homogeneousthroughout the whole arm

Scientific Programming 5

331 Absolute Orientation Filter Combining the measuredangular velocity acceleration and magnetic field values ofone single IMMU it can stably determine the orientationswith respect to a global coordinate system The globalcoordinate119873 and each coordinate 119887 of IMMUs are shown inFigure 1 The transformation between the representations ofa 3 times 1 column-vector x between119873 and 119887 is expressed as

x119887 = C119887119899 [q] x119899 (10)

where quaternion q = [1199020Q] [Q] is the antisymmetricmatrix given by

[Q] = [[[

0 1198763 minus1198762minus1198763 0 11987611198762 minus1198761 0]]] (11)

The attitude matrix C is related to the quaternion by

C (119902) = (11990220 minusQ sdotQ) I + 2QQ119879 + 21199020 [Q] (12)

where I is the identity matrixThe state vector is composed of the rotation quaternion

The state transition vector equation is

x119896+1 = 119902119896+1 = Φ (119879119904 120596119896) + 119908119896 = exp (Ω119896119879119904) 119902119896 + 119902119908119896119902119908119896 = minus1198791199042 Γ119896119902k119896 = minus1198791199042 [[119890119896times] + 1199024119896Iminus119890119879119896 ] 119892k119896 (13)

where the gyro measurement noise vector 119892k119896 is assumedsmall enough that a first-order approximation of the noisytransition matrix is possible

Then the process noise covariancematrix119876119896 will have thefollowing expression

119876119896 = (1198791199042 )2 Γ119896Γ119892Γ119879119896 (14)

The measurement model is constructed by stacking theaccelerometer and magnetometer measurement vectors

z119896+1 = [ a119896+1m119896+1]

= [C119887119899 (119902119896+1) 00 C119887119899 (119902119896+1)] [

gh] + [119886k119896+1119898k119896+1

] (15)

The covariance matrix of the measurement model R119896+1 is

R119896+1 = [119886R119896+1 00 119898R119896+1

] (16)

where the accelerometer and magnetometer measurementnoise 119886k119896+1 and

119898k119896+1 are uncorrelated zero-meanwhite noiseprocess and the covariancematrixes of which are 119886R119896+1 = 1205902119886Iand 119898R119896+1 = 1205902119898I respectively

Upper arm Forearm Palm Proximal Medial DistalN H

Arm Hand Finger

Y

Z X z3 z4 z5 z6x1

y1 y3 y4 y5 y6y2

z1 x2z2 x3 x6x5x4

L1 L2

Wearable position of the IMMUJoint

A1 A2 F3F1 F2L4 L5 L6L3

Figure 2 The model and frames of the arm-hand

Because of the nonlinear nature of (15) the EKF approachrequires that a first-order Taylor-Maclaurin expansion iscarried out around the current state estimation by computingthe Jacobian matrix

Η119896+1 = 120597120597x119896+1 z119896+110038161003816100381610038161003816100381610038161003816x119896+1=xminus119896+1 (17)

Then the orientations are estimated by the following EKFequations

Compute the a priori state estimate

xminus119896+1 = Φ (T119904120596119896) x119896 (18)

Compute the a priori error covariance matrix

Pminus119896+1 = Φ (T119904120596119896)P119896Φ(T119904120596119896)119879 +Q119896 (19)

Compute the Kalman gain

K119896+1 = Pminus119896+1H119879119896+1 (H119896+1Pminus119896+1H119879119896+1 + R119896+1)minus1 (20)

Compute the a posteriori state estimate

x119896+1 = xminus119896+1 + K119896+1 [z119896+1 minus 119891 (xminus119896+1)] (21)

Compute the a posteriori error covariance matrix

P119896+1 = Pminus119896+1 minus K119896+1H119896+1Pminus119896+1 (22)

According to the above algorithm the absolute orienta-tions of each IMMU can be estimated Then the kinematicsof the human arm and hand is considered

332 Relative Orientation Filter The kinematic frames ofthe arm hand and the forefinger are presented There aresix joints and the coordinate frames are built including theglobal coordinate (119873) upper-arm coordinate (1198601) forearm(1198602) palm coordinate (119867) proximal coordinate (1198651) medialcoordinate (1198652) and distal coordinate (1198653) And the lengthsbetween the joints are 1198711 1198712 1198713 1198714 1198715 and 1198716 The framesare shown in Figure 2Then the relative orientations betweentwo consecutive bodies can be determined by the following

q119903119894119895 = qminus1119894 sdot q119895 (23)

where q119903119894119895 is the quaternion of the relative orientationsq119894 is the quaternion of the absolute orientations of the

6 Scientific Programming

Gyros Accelerometers Magnetometers

Online calibration Offline calibration

Orientations estimation based QEKF

Constraint

Relative orientations

Absolute orientations

Kinematic parameters

Positions

Yes No

Kinematics of arm-hand

Figure 3 The diagram of the proposed method

first coordinate and q119895 is the quaternion of the absoluteorientations of the second coordinate

Meanwhile the kinematic models of the arm hand andfingers are considered to determine the orientations of eachsegment Human arm-hand motion can be supposed to bethe articulatedmotion of rigid body partsThese segments areupper arm (between the shoulder and elbow joints) forearm(between the elbow and wrist joints) hand (between thewrist joint and proximal joint) proximal finger (between theproximal joint and medial joint) medial finger (between themedial joint and distal joint) and distal finger (from the distaljoint) Every joint has its own local frame Shoulder can bemodeled as a ball joint with three DOFs and a fixed pointrepresenting the center of the shoulderMovements are repre-sented as the vector between the upper arm and body Elbowis the rotating hinge joint with two DOFs Wrist is modeledas a rotating hinge joint with two DOFs that is calculatedbetween the vector representing the hand and forearm Prox-imal joint is modeled as rotating hinge joint with two DOFsMedial joint and distal joint aremodeled as rotating joint withone DOF Thus the kinematic of this model consists of tenDOFs three in the shoulder joint two in the elbow joint twoin the wrist joint two in the proximal joint one in the medialjoint and one in the distal joint Hence the respondingconstraints are used to determine the orientations of eachsegment

34 Positions Estimation We assume the body keeps staticand the motion of arm and hand is formed by rotation of thejoints Hence the position of the fingertip p1198731198653 expressed in

the hand coordinate frame (see Figure 2) can be derived usingforward kinematics

[p11987311986531 ] = T1198731198601T11986011198602T1198602119867T1198671198651T11986511198652T11986521198653p11986531198652

= T1198731198653 [p119865311986521 ] (24)

where the transformation between two consecutive bodies isexpressed by T1198731198601 T11986011198602 T1198602119867 T1198671198651 T11986511198652 and T11986521198653

The total transformation T1198731198653 is given by the product ofeach consecutive contribution

T1198731198653 = [119877 (1199021198731198653) p119873119865301198793 1 ] (25)

where 119877(1199021198731198653) is the orientation of the distal phalanx withrespect to the body and p1198731198653 is the position of the distal frameexpressed in the global frame

35 Summary The gestures capture method is proposed inthe section The two procedures of the calibration are firstlyimplemented to improve the measurements of the sensorsThe inertial sensors are calibrated by the offline calibrationand the gyros are calibrated by the online calibration Thenthe orientations of the IMMUs are estimated by the QEKFAnd the kinematics of arm-hand is considered and the con-straints are integrated to determine the relative orientationsFinally the kinematic parameters are used to estimate thepositions A complete diagram of the method is depicted inFigure 3

Scientific Programming 7

minus20minus10

010

20

minus20minus10

010

20minus15

minus10

minus5

0

5

10

15z

(ms

2)

y (ms 2) x (ms2 )

Raw dataCalibrated

(a) The calibration results of accelerometers

minus1000minus500

0500

1000

minus1000minus500

0500

1000

minus1000

minus500

0

500

1000

Raw dataCalibrated

z(n

T)

y (nT) x (nT)

(b) The calibration results of magnetometers

Figure 4 The calibration results

Table 1 The calibration results

Scale Bias

Accelerometers G119886 = [[[[[

1 minus002 00 095 0120 0 092

]]]]]

b119886 = [[[[[

056087minus087

]]]]]

Magnetometers Gℎ = [[[[[

111 002 00 110 00 0 105

]]]]]

bℎ = [[[[[

571minus3954minus8298

]]]]]

4 Experiments and Results

The calibration is implemented to improve the sensors accu-racy and then the comparison experiments are testified toinvestigate the stability and accuracy of the orientations esti-mation The real-time gesture motion capture experimentsprove the validness of the proposed device

41 Calibration Results The calibration results of the triaxisaccelerometers and triaxis magnetometers are presented inthe section The calibration samples of IMMUs are collectedby rotating in various orientations Then the proposedmethod was used to determine the correctional calibrationparameters as G119886 Gℎ b119886 and bℎ The outputs of theaccelerometers and magnetometers are shown in Figure 4The blue sphere in Figure 4 is the raw date of sensors and thered sphere is the calibrated data of sensors The calibrationparameters are listed in Table 1

42 Evaluation Experiments After calibrating all the sensorsof the device the evaluation experiments are designed toassess the accuracy of the orientation estimation One of thestrings of the device is attached to the robotic arm ThreeIMMUs are set the same directions shown in Figure 5 Andthe robotic arm is rotated from the first state to the secondstate Then the estimated orientations of IMMUs compare

Table 2 The RMSE of the orientations

Static (∘) Dynamic (∘)Yaw Pitch Roll Yaw Pitch Roll

IMMU-1 004 001 002 243 005 102IMMU-2 011 001 002 168 051 103IMMU-3 010 002 002 177 002 007

with the true orientations of robotic arm The results of theorientations of three IMMUs are shown in Figure 6 Andthe results of the root mean square error (RMSE) of theorientations are listed in Table 2

As shown in Figure 6 during 5sim8 s the robotic armis dynamic And in static case it can be known that theresults have smaller variance and higher precision than theresults of the dynamic case Moreover the yaw angle has thelager variance compared with the roll and pitch It shouldbe caused by the motors of the robotic arm because themagnetometers are disturbed When the robotic arm is instatic the device can compute the accuracy orientations Thecomparison results proved the accuracy of the orientations ofthe IMMUs of the device Then the real-time gesture motioncapture experiments are implemented

43 Gesture Capture Experiments The IMMUsrsquo data is sam-pled collected and computed by the MCU and subsequentlytransmitted via Bluetooth to the external devices The MCUprocesses the raw data estimates the orientations of the eachunit encapsulates them into a packet and then sends thepacket to the PC by BluetoothThe baud rate for transmittingdata is 115200 bps The frequency is 50HZ By using thisdesign themotion capture can be demonstrated by the virtualmodel on the PC immediately The interface is written by CThe wearable system is shown as Figure 7

To further verify the effectiveness of proposed deviceupper arm is swung up and down The results of theorientations of the gestures are shown in Figure 8 And the

8 Scientific Programming

Robotic arm

IMMU-1

IMMU-2

IMMU-3

z1

x1

y1

(a) The first state

(b) The second state

Figure 5 The comparison experiments with robotic arm

positons of the fingertip of the right forefinger are shownin Figure 9 As shown in the figures the wearable devicecan determine the realistic movements of the arm andhand Meanwhile the accuracy of the results is assessed bythe statistics The root mean square error (RMSE) of theorientations is listed in Table 3 The RMSE of the positonsare listed in Table 4 Furthermore the wearable device isalso tested by ten healthy participants the real-time motiongestures capture experiments are implemented to prove thevalidity of the proposed system

44 Discussion In this paper a novel wearable device forgestures capturing based on magnetic and inertial measure-ment units is proposed As a practically useful applicationthe proposed device satisfies the requirements including theaccuracy computational efficiency and robustness First thecalibration method is designed to improve the accuracy of

0 5 10 15 20 25minus200

0

200

Time (s)

IMMU-1

YawPitch

Roll

YawPitch

Roll

YawPitch

Roll

IMMU-2

IMMU-3

Ang

el (∘

)

0 5 10 15 20 25minus200

0

200

Time (s)

Ang

el (∘

)0 5 10 15 20 25

minus200

0

200

Time (s)

Ang

el (∘

)

Figure 6 The results of orientations of the IMMUs

IMMU

Bluetooth

Data glovePC

Virtual modelTrajectory

Figure 7 The wearable system

Table 3 The RMSE of the orientations of the arm-hand

Angle (∘)Upper arm 107 110 044forearm 023 024Palm 019 018Proximal 011 026Medial 010Distal 008

Table 4 The RMSE of the positions of the fingertip

119909 (cm) 119910 (cm) 119911 (cm)Positions of the fingertip 345 135 234

the measurements of the sensors In particular the magneticeffects of the field are attenuated in advance Then the

Scientific Programming 9

0 10 20 30 40minus100

0

100Upper arm

t (s)0 10 20 30 40

minus1

0

1Forearm

t (s)

0 10 20 30 40minus1

0

1Palm

t (s)0 10 20 30 40

minus2

0

2Proximal

t (s)

0 10 20 30 40

0

05Medial

t (s)0 10 20 30 40

minus05minus05

0

05Distal

t (s)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Figure 8 The orientations of gestures

2040

6080 minus15 minus1 minus05 0 05 1

minus20

0

20

40

60

y (cm)x (cm)

z (c

m)

StartEnd

Figure 9 The positions of the fingertip of the right forefinger

orientations of gesture are divided into absolute orienta-tions and relative orientations The absolute orientations aredetermined by the QEKFs and the relative orientations areestimated by integrating the kinematics of the arm-handThe

positions are then easily computed The proposed method issimple and fast The algorithm is operated in the embeddedsystem at 50Hz The results of the experiments proved theadvantages of the proposed device

5 Conclusion

This paper presents the design implementation and exper-imental results of a wearable device for gestures captur-ing using inertial and magnetic sensor modules containingorthogonally mounted triads of accelerometers angular ratesensors and magnetometers Different from commercialmotion data gloves which usually use high-cost motion-sensing fibers to acquire hand motion data we adoptedthe low-cost inertial and magnetic sensor to reduce costMeanwhile the performance of the low-cost low-power andlight-weight IMMU is superior to some commercial IMMUFurthermore the novel device is proposed based on thirty-six IMMUs which cover the whole segments of the two armsand hands We designed the online and offline calibrationmethods to improve the accuracy of the units We deducedthe 3D arm and hand motion estimation algorithms thatintegrated the proposed kinematic models of the arm handand fingers and the attitude of gesture and positions offingertips can be determined For real-time performance and

10 Scientific Programming

convenience the interface with virtual model is designedPerformance evaluations verified that the proposed dataglove can accurately capture the motion of gestures Thesystem is developed in the way that all electronic componentscan be integrated and easy to wear This makes it moreconvenient and appealing for the user

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work has been supported by the National Science Foun-dation for Young Scientists of China (Grant no 61503212) andNational Natural Science Foundation of China (Grant nosU1613212 61327809 61210013 and 91420302)

References

[1] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) pp 25ndash28 Kitakyushu Japan August 2014

[2] J M Palacios C Sagues E Montijano and S LlorenteldquoHuman-computer interaction based on hand gestures usingRGB-D sensorsrdquo Sensors vol 13 no 9 pp 11842ndash11860 2013

[3] S S Rautaray and A Agrawal ldquoVision based hand gesturerecognition for human computer interaction a surveyrdquo Artifi-cial Intelligence Review vol 43 no 1 pp 1ndash54 2012

[4] E A Arkenbout J C F de Winter and P Breedveld ldquoRobusthandmotion tracking through data fusion of 5dt data glove andnimble VR kinect camera measurementsrdquo Sensors vol 15 no12 pp 31644ndash31671 2015

[5] D Regazzoni G De Vecchi and C Rizzi ldquoRGB cams vs RGB-D sensors low cost motion capture technologies performancesand limitationsrdquo Journal of Manufacturing Systems vol 33 no4 pp 719ndash728 2014

[6] Y Li X Chen X Zhang K Wang and Z J Wang ldquoA sign-component-based framework for Chinese sign language recog-nition using accelerometer and sEMG datardquo IEEE Transactionson Biomedical Engineering vol 59 no 10 pp 2695ndash2704 2012

[7] L Dipietro A M Sabatini and P Dario ldquoA survey of glove-based systems and their applicationsrdquo IEEE Transactions onSystems Man and Cybernetics Part C Applications and Reviewsvol 38 no 4 pp 461ndash482 2008

[8] B Takacs ldquoHow and why affordable virtual reality shapes thefuture of educationrdquoThe International Journal of Virtual Realityvol 7 no 1 pp 53ndash66 2008

[9] G Saggio ldquoA novel array of flex sensors for a goniometric gloverdquoSensors and Actuators A Physical vol 205 pp 119ndash125 2014

[10] R Gentner and J Classen ldquoDevelopment and evaluation of alow-cost sensor glove for assessment of human finger move-ments in neurophysiological settingsrdquo Journal of NeuroscienceMethods vol 178 no 1 pp 138ndash147 2009

[11] X L Wang and C L Yang ldquoConstructing gyro-free inertialmeasurement unit from dual accelerometers for gesture detec-tionrdquo Sensors amp Transducers vol 171 no 5 pp 134ndash140 2014

[12] W Chou B Fang L Ding X Ma and X Guo ldquoTwo-stepoptimal filter design for the low-cost attitude and headingreference systemsrdquo IET Science Measurement and Technologyvol 7 no 4 pp 240ndash248 2013

[13] D Roetenberg H J Luinge C T M Baten and P H VeltinkldquoCompensation of magnetic disturbances improves inertial andmagnetic sensing of human body segment orientationrdquo IEEETransactions on Neural Systems and Rehabilitation Engineeringvol 13 no 3 pp 395ndash405 2005

[14] J M Lambrecht and R F Kirsch ldquoMiniature low-powerinertial sensors promising technology for implantable motioncapture systemsrdquo IEEE Transactions on Neural Systems andRehabilitation Engineering vol 22 no 6 pp 1138ndash1147 2014

[15] R Xu S L Zhou and W J Li ldquoMEMS accelerometer basednonspecific-user hand gesture recognitionrdquo IEEE Sensors Jour-nal vol 12 no 5 pp 1166ndash1173 2012

[16] E Morganti L Angelini A Adami D Lalanne L Lorenzelliand E Mugellini ldquoA smart watch with embedded sensorsto recognize objects grasps and forearm gesturesrdquo ProcediaEngineering vol 41 pp 1169ndash1175 2012

[17] J-H Kim N DThang and T-S Kim ldquo3-D handmotion track-ing and gesture recognition using a data gloverdquo in Proceedingsof the IEEE International Symposium on Industrial Electronics(ISIE rsquo09) pp 1013ndash1018 IEEE Seoul South Korea July 2009

[18] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) vol 10 pp 25ndash28 Kitakyushu Japan August2014

[19] F Cavallo D Esposito E Rovini et al ldquoPreliminary evalu-ation of SensHand V1 in assessing motor skills performancein Parkinson diseaserdquo in Proceedings of the 2013 IEEE 13thInternational Conference onRehabilitation Robotics (ICORR rsquo13)pp 1ndash6 Seattle Wash USA June 2013

[20] H G Kortier V I Sluiter D Roetenberg and P H VeltinkldquoAssessment of hand kinematics using inertial and magneticsensorsrdquo Journal of NeuroEngineering and Rehabilitation vol 11no 1 article 70 2014

[21] Y Xu Y Wang Y Su and X Zhu ldquoResearch on the calibrationmethod of micro inertial measurement unit for engineeringapplicationrdquo Journal of Sensors vol 2016 Article ID 9108197 11pages 2016

[22] J Keighobadi ldquoFuzzy calibration of a magnetic compass forvehicular applicationsrdquoMechanical Systems and Signal Process-ing vol 25 no 6 pp 1973ndash1987 2011

[23] X Yun and E R Bachmann ldquoDesign implementation andexperimental results of a quaternion-based Kalman filter forhuman body motion trackingrdquo IEEE Transactions on Roboticsvol 22 no 6 pp 1216ndash1227 2006

[24] Xsens Technologies httpwwwxsenscom[25] T Hamel and R Mahony ldquoAttitude estimation on SO(3) based

on direct inertial measurementsrdquo in Proceedings of the IEEEInternational Conference on Robotics and Automation (ICRArsquo06) pp 2170ndash2175 Orlando Fla USA May 2006

[26] S Bonnabel PMartin and P Rouchon ldquoNon-linear symmetry-preserving observers on Lie groupsrdquo IEEE Transactions onAutomatic Control vol 54 no 7 pp 1709ndash1713 2009

[27] InvenSense MPU-9250 Nine-Axis MEMS MotionTrackingDevice 2015 httpwwwinvensensecom-productsmotion-tracking9-axismpu-9250

4 Scientific Programming

where Mℎ is a 3 times 3 matrix that scales the magnetometersoutputs bℎ denotes the disturbance vector including magne-tometersrsquo bias and magnetic effects andwℎ isin R3 is the noisesthat supposed to be Gaussian with zero-means

According to the above presentations we can know thatthe models of accelerometers and magnetometers includeerrors which would lead to errors in estimating orientationsHence the calibration should be implemented to improve theaccuracy of the sensors

32 Calibration In order to improve the accuracy of theIMMU calibration is a necessary procedure The typicalcalibration method is to assign the inertial sensors to aknown angular velocity and linear acceleration This methodnormally needs some specific equipment such as turntableHere a novel calibration method is proposed that do notneed the specific equipment and is easy to implement Thecalibration procedure has two steps First the accelerometersand magnetometers are calibrated by the offline procedureThen the gyro is calibrated by the online procedure

The unifiedmathematical models of the calibration of theaccelerometers and magnetometers can be used as follows

h119887119898 = Mℎh119887 + bℎ + w119898a119887119898 = M119886a119887 + b119886 + w119886 (5)

where triaxial sensor model is written in the vector form andbℎ b119886 are constant offset that shift the outputs of sensors

The calibration is the process that determines coefficientsbℎ b119886 Mℎ M119886 to improve the measurements of sensorsWhen the unit has omnidirectional rotation the magnitudesof the true magnetic field and gravity field remain constantand the loci of the true magnetic field measured h119887 and a119887 arespherical Meanwhile the measured h119887119898 and a

119887119898 are ellipsoids

and they can be expressed as follows [31]10038171003817100381710038171003817h119887100381710038171003817100381710038172 = (h119887119898)119879Aℎh119887119898 minus 2 (bℎ)119879Aℎh119887119898 + (bℎ)119879Aℎbℎ

+ w11989810038171003817100381710038171003817a119887100381710038171003817100381710038172 = (a119887119898)119879A119886a119887119898 minus 2 (b119886)119879A119886a119887119898 + (b119886)119879A119886b119886

+ w119886(6)

where Aℎ = (Gℎ)119879Gℎ Gℎ = (Mℎ)minus1 A119886 = (G119886)119879G119886 G119886 =(M119886)minus1 and w119898 and w119886 are assumed noiseEquations (6) are the expressions of the ellipsoid In

other words the measurements are constrained to lie on anellipsoid Thus the calibration of the magnetometers andaccelerometers is to seek ellipsoid-fittingmethods to solve thecoefficients ofG119886Gℎ b119886 bℎ And the least squares algorithmis commonly used to determine the parameters The deviceis rotated adequately to ensure that each unit gets enoughmeasurements to determine the calibration parameters byleast squares algorithm Hence the accelerometers and mag-netometers of all units of the device are entirely calibrated

After calibration the magnetometers are calibrated tolessen the environmental magnetic effect and the biases

of the sensor outputs are compensated Accordingly theprevious equation can be rewritten as follows

h119898 = C119887119899Hℎ + wℎ (7)

Since the scale and bias errors from the accelerometersare calibrated the models of the accelerometers rewritten asfollows

a119898 = C119887119899g + w119886 (8)

The offline calibration is implemented to determine thebias and scale of the accelerometers andmagnetometers Andthe online calibration is implemented to remove the gyro biasWe keep the data glove stationary for a while before use Sothe readings are zeros The bias can then be computed by themean value of the measurements The model is expressed asfollows

120596119898 = 120596 + w119892 (9)

where 120596119898 are the measurements of gyros 120596 are the trueangular velocity and w119892 are the noise of gyros

The calibration parameters of the sensors are compen-sated into the measurements so that the accuracy can beimproved for further work

33 Orientation Filters Based on these kinds of sensors twoindependent ways can determine the attitude and headingOne is determined by open-loop gyros The angular rate ofthe rigid body is measured using gyros with respect to itsbody axis frame The angles are estimated by the open-loopintegration process which has high dynamic characteristicHowever the gyro errors would cause wandering attitudeangles and the gradual instability of the integration driftingThe other way is determined from open-loop accelerome-ters and magnetometers The orientations can be correctlyobtained from accelerometers and magnetometers in theideal environment However the disturbance and noiseswould lead to large errors and make results lack reliabilityThe two ways are both quite difficult to achieve acceptableperformance Sensor fusion is a great choice to attain thestable and accurate orientations Then the fusion algorithmof quaternion-based extended Kalman filter will be deduced

The frames are shown in Figure 1 In the figure the globalframe is 119873 and local reference frame is respectively locatedin each IMMU The global reference frame 119911-axis is definedalong the axial axis (from the head to the feet) of the subject119910-axis along the sagittal (from the left shoulder to the rightshoulder) axis and 119909-axis along the coronal axis (from theback to the chest) The local frame 119911-axis is defined alongthe axial axis (normal to the surface of IMMU along thedownward) of the subject 119910-axis along the sagittal (fromthe left side to the right side of the IMMU) axis and 119909-axis along the coronal axis (from the back to the forward ofIMMU)Meanwhile two assumptions about data glove in usearemade (1) the body keeps static and only arm and hand arein motion (2) the local static magnetic field is homogeneousthroughout the whole arm

Scientific Programming 5

331 Absolute Orientation Filter Combining the measuredangular velocity acceleration and magnetic field values ofone single IMMU it can stably determine the orientationswith respect to a global coordinate system The globalcoordinate119873 and each coordinate 119887 of IMMUs are shown inFigure 1 The transformation between the representations ofa 3 times 1 column-vector x between119873 and 119887 is expressed as

x119887 = C119887119899 [q] x119899 (10)

where quaternion q = [1199020Q] [Q] is the antisymmetricmatrix given by

[Q] = [[[

0 1198763 minus1198762minus1198763 0 11987611198762 minus1198761 0]]] (11)

The attitude matrix C is related to the quaternion by

C (119902) = (11990220 minusQ sdotQ) I + 2QQ119879 + 21199020 [Q] (12)

where I is the identity matrixThe state vector is composed of the rotation quaternion

The state transition vector equation is

x119896+1 = 119902119896+1 = Φ (119879119904 120596119896) + 119908119896 = exp (Ω119896119879119904) 119902119896 + 119902119908119896119902119908119896 = minus1198791199042 Γ119896119902k119896 = minus1198791199042 [[119890119896times] + 1199024119896Iminus119890119879119896 ] 119892k119896 (13)

where the gyro measurement noise vector 119892k119896 is assumedsmall enough that a first-order approximation of the noisytransition matrix is possible

Then the process noise covariancematrix119876119896 will have thefollowing expression

119876119896 = (1198791199042 )2 Γ119896Γ119892Γ119879119896 (14)

The measurement model is constructed by stacking theaccelerometer and magnetometer measurement vectors

z119896+1 = [ a119896+1m119896+1]

= [C119887119899 (119902119896+1) 00 C119887119899 (119902119896+1)] [

gh] + [119886k119896+1119898k119896+1

] (15)

The covariance matrix of the measurement model R119896+1 is

R119896+1 = [119886R119896+1 00 119898R119896+1

] (16)

where the accelerometer and magnetometer measurementnoise 119886k119896+1 and

119898k119896+1 are uncorrelated zero-meanwhite noiseprocess and the covariancematrixes of which are 119886R119896+1 = 1205902119886Iand 119898R119896+1 = 1205902119898I respectively

Upper arm Forearm Palm Proximal Medial DistalN H

Arm Hand Finger

Y

Z X z3 z4 z5 z6x1

y1 y3 y4 y5 y6y2

z1 x2z2 x3 x6x5x4

L1 L2

Wearable position of the IMMUJoint

A1 A2 F3F1 F2L4 L5 L6L3

Figure 2 The model and frames of the arm-hand

Because of the nonlinear nature of (15) the EKF approachrequires that a first-order Taylor-Maclaurin expansion iscarried out around the current state estimation by computingthe Jacobian matrix

Η119896+1 = 120597120597x119896+1 z119896+110038161003816100381610038161003816100381610038161003816x119896+1=xminus119896+1 (17)

Then the orientations are estimated by the following EKFequations

Compute the a priori state estimate

xminus119896+1 = Φ (T119904120596119896) x119896 (18)

Compute the a priori error covariance matrix

Pminus119896+1 = Φ (T119904120596119896)P119896Φ(T119904120596119896)119879 +Q119896 (19)

Compute the Kalman gain

K119896+1 = Pminus119896+1H119879119896+1 (H119896+1Pminus119896+1H119879119896+1 + R119896+1)minus1 (20)

Compute the a posteriori state estimate

x119896+1 = xminus119896+1 + K119896+1 [z119896+1 minus 119891 (xminus119896+1)] (21)

Compute the a posteriori error covariance matrix

P119896+1 = Pminus119896+1 minus K119896+1H119896+1Pminus119896+1 (22)

According to the above algorithm the absolute orienta-tions of each IMMU can be estimated Then the kinematicsof the human arm and hand is considered

332 Relative Orientation Filter The kinematic frames ofthe arm hand and the forefinger are presented There aresix joints and the coordinate frames are built including theglobal coordinate (119873) upper-arm coordinate (1198601) forearm(1198602) palm coordinate (119867) proximal coordinate (1198651) medialcoordinate (1198652) and distal coordinate (1198653) And the lengthsbetween the joints are 1198711 1198712 1198713 1198714 1198715 and 1198716 The framesare shown in Figure 2Then the relative orientations betweentwo consecutive bodies can be determined by the following

q119903119894119895 = qminus1119894 sdot q119895 (23)

where q119903119894119895 is the quaternion of the relative orientationsq119894 is the quaternion of the absolute orientations of the

6 Scientific Programming

Gyros Accelerometers Magnetometers

Online calibration Offline calibration

Orientations estimation based QEKF

Constraint

Relative orientations

Absolute orientations

Kinematic parameters

Positions

Yes No

Kinematics of arm-hand

Figure 3 The diagram of the proposed method

first coordinate and q119895 is the quaternion of the absoluteorientations of the second coordinate

Meanwhile the kinematic models of the arm hand andfingers are considered to determine the orientations of eachsegment Human arm-hand motion can be supposed to bethe articulatedmotion of rigid body partsThese segments areupper arm (between the shoulder and elbow joints) forearm(between the elbow and wrist joints) hand (between thewrist joint and proximal joint) proximal finger (between theproximal joint and medial joint) medial finger (between themedial joint and distal joint) and distal finger (from the distaljoint) Every joint has its own local frame Shoulder can bemodeled as a ball joint with three DOFs and a fixed pointrepresenting the center of the shoulderMovements are repre-sented as the vector between the upper arm and body Elbowis the rotating hinge joint with two DOFs Wrist is modeledas a rotating hinge joint with two DOFs that is calculatedbetween the vector representing the hand and forearm Prox-imal joint is modeled as rotating hinge joint with two DOFsMedial joint and distal joint aremodeled as rotating joint withone DOF Thus the kinematic of this model consists of tenDOFs three in the shoulder joint two in the elbow joint twoin the wrist joint two in the proximal joint one in the medialjoint and one in the distal joint Hence the respondingconstraints are used to determine the orientations of eachsegment

34 Positions Estimation We assume the body keeps staticand the motion of arm and hand is formed by rotation of thejoints Hence the position of the fingertip p1198731198653 expressed in

the hand coordinate frame (see Figure 2) can be derived usingforward kinematics

[p11987311986531 ] = T1198731198601T11986011198602T1198602119867T1198671198651T11986511198652T11986521198653p11986531198652

= T1198731198653 [p119865311986521 ] (24)

where the transformation between two consecutive bodies isexpressed by T1198731198601 T11986011198602 T1198602119867 T1198671198651 T11986511198652 and T11986521198653

The total transformation T1198731198653 is given by the product ofeach consecutive contribution

T1198731198653 = [119877 (1199021198731198653) p119873119865301198793 1 ] (25)

where 119877(1199021198731198653) is the orientation of the distal phalanx withrespect to the body and p1198731198653 is the position of the distal frameexpressed in the global frame

35 Summary The gestures capture method is proposed inthe section The two procedures of the calibration are firstlyimplemented to improve the measurements of the sensorsThe inertial sensors are calibrated by the offline calibrationand the gyros are calibrated by the online calibration Thenthe orientations of the IMMUs are estimated by the QEKFAnd the kinematics of arm-hand is considered and the con-straints are integrated to determine the relative orientationsFinally the kinematic parameters are used to estimate thepositions A complete diagram of the method is depicted inFigure 3

Scientific Programming 7

minus20minus10

010

20

minus20minus10

010

20minus15

minus10

minus5

0

5

10

15z

(ms

2)

y (ms 2) x (ms2 )

Raw dataCalibrated

(a) The calibration results of accelerometers

minus1000minus500

0500

1000

minus1000minus500

0500

1000

minus1000

minus500

0

500

1000

Raw dataCalibrated

z(n

T)

y (nT) x (nT)

(b) The calibration results of magnetometers

Figure 4 The calibration results

Table 1 The calibration results

Scale Bias

Accelerometers G119886 = [[[[[

1 minus002 00 095 0120 0 092

]]]]]

b119886 = [[[[[

056087minus087

]]]]]

Magnetometers Gℎ = [[[[[

111 002 00 110 00 0 105

]]]]]

bℎ = [[[[[

571minus3954minus8298

]]]]]

4 Experiments and Results

The calibration is implemented to improve the sensors accu-racy and then the comparison experiments are testified toinvestigate the stability and accuracy of the orientations esti-mation The real-time gesture motion capture experimentsprove the validness of the proposed device

41 Calibration Results The calibration results of the triaxisaccelerometers and triaxis magnetometers are presented inthe section The calibration samples of IMMUs are collectedby rotating in various orientations Then the proposedmethod was used to determine the correctional calibrationparameters as G119886 Gℎ b119886 and bℎ The outputs of theaccelerometers and magnetometers are shown in Figure 4The blue sphere in Figure 4 is the raw date of sensors and thered sphere is the calibrated data of sensors The calibrationparameters are listed in Table 1

42 Evaluation Experiments After calibrating all the sensorsof the device the evaluation experiments are designed toassess the accuracy of the orientation estimation One of thestrings of the device is attached to the robotic arm ThreeIMMUs are set the same directions shown in Figure 5 Andthe robotic arm is rotated from the first state to the secondstate Then the estimated orientations of IMMUs compare

Table 2 The RMSE of the orientations

Static (∘) Dynamic (∘)Yaw Pitch Roll Yaw Pitch Roll

IMMU-1 004 001 002 243 005 102IMMU-2 011 001 002 168 051 103IMMU-3 010 002 002 177 002 007

with the true orientations of robotic arm The results of theorientations of three IMMUs are shown in Figure 6 Andthe results of the root mean square error (RMSE) of theorientations are listed in Table 2

As shown in Figure 6 during 5sim8 s the robotic armis dynamic And in static case it can be known that theresults have smaller variance and higher precision than theresults of the dynamic case Moreover the yaw angle has thelager variance compared with the roll and pitch It shouldbe caused by the motors of the robotic arm because themagnetometers are disturbed When the robotic arm is instatic the device can compute the accuracy orientations Thecomparison results proved the accuracy of the orientations ofthe IMMUs of the device Then the real-time gesture motioncapture experiments are implemented

43 Gesture Capture Experiments The IMMUsrsquo data is sam-pled collected and computed by the MCU and subsequentlytransmitted via Bluetooth to the external devices The MCUprocesses the raw data estimates the orientations of the eachunit encapsulates them into a packet and then sends thepacket to the PC by BluetoothThe baud rate for transmittingdata is 115200 bps The frequency is 50HZ By using thisdesign themotion capture can be demonstrated by the virtualmodel on the PC immediately The interface is written by CThe wearable system is shown as Figure 7

To further verify the effectiveness of proposed deviceupper arm is swung up and down The results of theorientations of the gestures are shown in Figure 8 And the

8 Scientific Programming

Robotic arm

IMMU-1

IMMU-2

IMMU-3

z1

x1

y1

(a) The first state

(b) The second state

Figure 5 The comparison experiments with robotic arm

positons of the fingertip of the right forefinger are shownin Figure 9 As shown in the figures the wearable devicecan determine the realistic movements of the arm andhand Meanwhile the accuracy of the results is assessed bythe statistics The root mean square error (RMSE) of theorientations is listed in Table 3 The RMSE of the positonsare listed in Table 4 Furthermore the wearable device isalso tested by ten healthy participants the real-time motiongestures capture experiments are implemented to prove thevalidity of the proposed system

44 Discussion In this paper a novel wearable device forgestures capturing based on magnetic and inertial measure-ment units is proposed As a practically useful applicationthe proposed device satisfies the requirements including theaccuracy computational efficiency and robustness First thecalibration method is designed to improve the accuracy of

0 5 10 15 20 25minus200

0

200

Time (s)

IMMU-1

YawPitch

Roll

YawPitch

Roll

YawPitch

Roll

IMMU-2

IMMU-3

Ang

el (∘

)

0 5 10 15 20 25minus200

0

200

Time (s)

Ang

el (∘

)0 5 10 15 20 25

minus200

0

200

Time (s)

Ang

el (∘

)

Figure 6 The results of orientations of the IMMUs

IMMU

Bluetooth

Data glovePC

Virtual modelTrajectory

Figure 7 The wearable system

Table 3 The RMSE of the orientations of the arm-hand

Angle (∘)Upper arm 107 110 044forearm 023 024Palm 019 018Proximal 011 026Medial 010Distal 008

Table 4 The RMSE of the positions of the fingertip

119909 (cm) 119910 (cm) 119911 (cm)Positions of the fingertip 345 135 234

the measurements of the sensors In particular the magneticeffects of the field are attenuated in advance Then the

Scientific Programming 9

0 10 20 30 40minus100

0

100Upper arm

t (s)0 10 20 30 40

minus1

0

1Forearm

t (s)

0 10 20 30 40minus1

0

1Palm

t (s)0 10 20 30 40

minus2

0

2Proximal

t (s)

0 10 20 30 40

0

05Medial

t (s)0 10 20 30 40

minus05minus05

0

05Distal

t (s)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Figure 8 The orientations of gestures

2040

6080 minus15 minus1 minus05 0 05 1

minus20

0

20

40

60

y (cm)x (cm)

z (c

m)

StartEnd

Figure 9 The positions of the fingertip of the right forefinger

orientations of gesture are divided into absolute orienta-tions and relative orientations The absolute orientations aredetermined by the QEKFs and the relative orientations areestimated by integrating the kinematics of the arm-handThe

positions are then easily computed The proposed method issimple and fast The algorithm is operated in the embeddedsystem at 50Hz The results of the experiments proved theadvantages of the proposed device

5 Conclusion

This paper presents the design implementation and exper-imental results of a wearable device for gestures captur-ing using inertial and magnetic sensor modules containingorthogonally mounted triads of accelerometers angular ratesensors and magnetometers Different from commercialmotion data gloves which usually use high-cost motion-sensing fibers to acquire hand motion data we adoptedthe low-cost inertial and magnetic sensor to reduce costMeanwhile the performance of the low-cost low-power andlight-weight IMMU is superior to some commercial IMMUFurthermore the novel device is proposed based on thirty-six IMMUs which cover the whole segments of the two armsand hands We designed the online and offline calibrationmethods to improve the accuracy of the units We deducedthe 3D arm and hand motion estimation algorithms thatintegrated the proposed kinematic models of the arm handand fingers and the attitude of gesture and positions offingertips can be determined For real-time performance and

10 Scientific Programming

convenience the interface with virtual model is designedPerformance evaluations verified that the proposed dataglove can accurately capture the motion of gestures Thesystem is developed in the way that all electronic componentscan be integrated and easy to wear This makes it moreconvenient and appealing for the user

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work has been supported by the National Science Foun-dation for Young Scientists of China (Grant no 61503212) andNational Natural Science Foundation of China (Grant nosU1613212 61327809 61210013 and 91420302)

References

[1] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) pp 25ndash28 Kitakyushu Japan August 2014

[2] J M Palacios C Sagues E Montijano and S LlorenteldquoHuman-computer interaction based on hand gestures usingRGB-D sensorsrdquo Sensors vol 13 no 9 pp 11842ndash11860 2013

[3] S S Rautaray and A Agrawal ldquoVision based hand gesturerecognition for human computer interaction a surveyrdquo Artifi-cial Intelligence Review vol 43 no 1 pp 1ndash54 2012

[4] E A Arkenbout J C F de Winter and P Breedveld ldquoRobusthandmotion tracking through data fusion of 5dt data glove andnimble VR kinect camera measurementsrdquo Sensors vol 15 no12 pp 31644ndash31671 2015

[5] D Regazzoni G De Vecchi and C Rizzi ldquoRGB cams vs RGB-D sensors low cost motion capture technologies performancesand limitationsrdquo Journal of Manufacturing Systems vol 33 no4 pp 719ndash728 2014

[6] Y Li X Chen X Zhang K Wang and Z J Wang ldquoA sign-component-based framework for Chinese sign language recog-nition using accelerometer and sEMG datardquo IEEE Transactionson Biomedical Engineering vol 59 no 10 pp 2695ndash2704 2012

[7] L Dipietro A M Sabatini and P Dario ldquoA survey of glove-based systems and their applicationsrdquo IEEE Transactions onSystems Man and Cybernetics Part C Applications and Reviewsvol 38 no 4 pp 461ndash482 2008

[8] B Takacs ldquoHow and why affordable virtual reality shapes thefuture of educationrdquoThe International Journal of Virtual Realityvol 7 no 1 pp 53ndash66 2008

[9] G Saggio ldquoA novel array of flex sensors for a goniometric gloverdquoSensors and Actuators A Physical vol 205 pp 119ndash125 2014

[10] R Gentner and J Classen ldquoDevelopment and evaluation of alow-cost sensor glove for assessment of human finger move-ments in neurophysiological settingsrdquo Journal of NeuroscienceMethods vol 178 no 1 pp 138ndash147 2009

[11] X L Wang and C L Yang ldquoConstructing gyro-free inertialmeasurement unit from dual accelerometers for gesture detec-tionrdquo Sensors amp Transducers vol 171 no 5 pp 134ndash140 2014

[12] W Chou B Fang L Ding X Ma and X Guo ldquoTwo-stepoptimal filter design for the low-cost attitude and headingreference systemsrdquo IET Science Measurement and Technologyvol 7 no 4 pp 240ndash248 2013

[13] D Roetenberg H J Luinge C T M Baten and P H VeltinkldquoCompensation of magnetic disturbances improves inertial andmagnetic sensing of human body segment orientationrdquo IEEETransactions on Neural Systems and Rehabilitation Engineeringvol 13 no 3 pp 395ndash405 2005

[14] J M Lambrecht and R F Kirsch ldquoMiniature low-powerinertial sensors promising technology for implantable motioncapture systemsrdquo IEEE Transactions on Neural Systems andRehabilitation Engineering vol 22 no 6 pp 1138ndash1147 2014

[15] R Xu S L Zhou and W J Li ldquoMEMS accelerometer basednonspecific-user hand gesture recognitionrdquo IEEE Sensors Jour-nal vol 12 no 5 pp 1166ndash1173 2012

[16] E Morganti L Angelini A Adami D Lalanne L Lorenzelliand E Mugellini ldquoA smart watch with embedded sensorsto recognize objects grasps and forearm gesturesrdquo ProcediaEngineering vol 41 pp 1169ndash1175 2012

[17] J-H Kim N DThang and T-S Kim ldquo3-D handmotion track-ing and gesture recognition using a data gloverdquo in Proceedingsof the IEEE International Symposium on Industrial Electronics(ISIE rsquo09) pp 1013ndash1018 IEEE Seoul South Korea July 2009

[18] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) vol 10 pp 25ndash28 Kitakyushu Japan August2014

[19] F Cavallo D Esposito E Rovini et al ldquoPreliminary evalu-ation of SensHand V1 in assessing motor skills performancein Parkinson diseaserdquo in Proceedings of the 2013 IEEE 13thInternational Conference onRehabilitation Robotics (ICORR rsquo13)pp 1ndash6 Seattle Wash USA June 2013

[20] H G Kortier V I Sluiter D Roetenberg and P H VeltinkldquoAssessment of hand kinematics using inertial and magneticsensorsrdquo Journal of NeuroEngineering and Rehabilitation vol 11no 1 article 70 2014

[21] Y Xu Y Wang Y Su and X Zhu ldquoResearch on the calibrationmethod of micro inertial measurement unit for engineeringapplicationrdquo Journal of Sensors vol 2016 Article ID 9108197 11pages 2016

[22] J Keighobadi ldquoFuzzy calibration of a magnetic compass forvehicular applicationsrdquoMechanical Systems and Signal Process-ing vol 25 no 6 pp 1973ndash1987 2011

[23] X Yun and E R Bachmann ldquoDesign implementation andexperimental results of a quaternion-based Kalman filter forhuman body motion trackingrdquo IEEE Transactions on Roboticsvol 22 no 6 pp 1216ndash1227 2006

[24] Xsens Technologies httpwwwxsenscom[25] T Hamel and R Mahony ldquoAttitude estimation on SO(3) based

on direct inertial measurementsrdquo in Proceedings of the IEEEInternational Conference on Robotics and Automation (ICRArsquo06) pp 2170ndash2175 Orlando Fla USA May 2006

[26] S Bonnabel PMartin and P Rouchon ldquoNon-linear symmetry-preserving observers on Lie groupsrdquo IEEE Transactions onAutomatic Control vol 54 no 7 pp 1709ndash1713 2009

[27] InvenSense MPU-9250 Nine-Axis MEMS MotionTrackingDevice 2015 httpwwwinvensensecom-productsmotion-tracking9-axismpu-9250

Scientific Programming 5

331 Absolute Orientation Filter Combining the measuredangular velocity acceleration and magnetic field values ofone single IMMU it can stably determine the orientationswith respect to a global coordinate system The globalcoordinate119873 and each coordinate 119887 of IMMUs are shown inFigure 1 The transformation between the representations ofa 3 times 1 column-vector x between119873 and 119887 is expressed as

x119887 = C119887119899 [q] x119899 (10)

where quaternion q = [1199020Q] [Q] is the antisymmetricmatrix given by

[Q] = [[[

0 1198763 minus1198762minus1198763 0 11987611198762 minus1198761 0]]] (11)

The attitude matrix C is related to the quaternion by

C (119902) = (11990220 minusQ sdotQ) I + 2QQ119879 + 21199020 [Q] (12)

where I is the identity matrixThe state vector is composed of the rotation quaternion

The state transition vector equation is

x119896+1 = 119902119896+1 = Φ (119879119904 120596119896) + 119908119896 = exp (Ω119896119879119904) 119902119896 + 119902119908119896119902119908119896 = minus1198791199042 Γ119896119902k119896 = minus1198791199042 [[119890119896times] + 1199024119896Iminus119890119879119896 ] 119892k119896 (13)

where the gyro measurement noise vector 119892k119896 is assumedsmall enough that a first-order approximation of the noisytransition matrix is possible

Then the process noise covariancematrix119876119896 will have thefollowing expression

119876119896 = (1198791199042 )2 Γ119896Γ119892Γ119879119896 (14)

The measurement model is constructed by stacking theaccelerometer and magnetometer measurement vectors

z119896+1 = [ a119896+1m119896+1]

= [C119887119899 (119902119896+1) 00 C119887119899 (119902119896+1)] [

gh] + [119886k119896+1119898k119896+1

] (15)

The covariance matrix of the measurement model R119896+1 is

R119896+1 = [119886R119896+1 00 119898R119896+1

] (16)

where the accelerometer and magnetometer measurementnoise 119886k119896+1 and

119898k119896+1 are uncorrelated zero-meanwhite noiseprocess and the covariancematrixes of which are 119886R119896+1 = 1205902119886Iand 119898R119896+1 = 1205902119898I respectively

Upper arm Forearm Palm Proximal Medial DistalN H

Arm Hand Finger

Y

Z X z3 z4 z5 z6x1

y1 y3 y4 y5 y6y2

z1 x2z2 x3 x6x5x4

L1 L2

Wearable position of the IMMUJoint

A1 A2 F3F1 F2L4 L5 L6L3

Figure 2 The model and frames of the arm-hand

Because of the nonlinear nature of (15) the EKF approachrequires that a first-order Taylor-Maclaurin expansion iscarried out around the current state estimation by computingthe Jacobian matrix

Η119896+1 = 120597120597x119896+1 z119896+110038161003816100381610038161003816100381610038161003816x119896+1=xminus119896+1 (17)

Then the orientations are estimated by the following EKFequations

Compute the a priori state estimate

xminus119896+1 = Φ (T119904120596119896) x119896 (18)

Compute the a priori error covariance matrix

Pminus119896+1 = Φ (T119904120596119896)P119896Φ(T119904120596119896)119879 +Q119896 (19)

Compute the Kalman gain

K119896+1 = Pminus119896+1H119879119896+1 (H119896+1Pminus119896+1H119879119896+1 + R119896+1)minus1 (20)

Compute the a posteriori state estimate

x119896+1 = xminus119896+1 + K119896+1 [z119896+1 minus 119891 (xminus119896+1)] (21)

Compute the a posteriori error covariance matrix

P119896+1 = Pminus119896+1 minus K119896+1H119896+1Pminus119896+1 (22)

According to the above algorithm the absolute orienta-tions of each IMMU can be estimated Then the kinematicsof the human arm and hand is considered

332 Relative Orientation Filter The kinematic frames ofthe arm hand and the forefinger are presented There aresix joints and the coordinate frames are built including theglobal coordinate (119873) upper-arm coordinate (1198601) forearm(1198602) palm coordinate (119867) proximal coordinate (1198651) medialcoordinate (1198652) and distal coordinate (1198653) And the lengthsbetween the joints are 1198711 1198712 1198713 1198714 1198715 and 1198716 The framesare shown in Figure 2Then the relative orientations betweentwo consecutive bodies can be determined by the following

q119903119894119895 = qminus1119894 sdot q119895 (23)

where q119903119894119895 is the quaternion of the relative orientationsq119894 is the quaternion of the absolute orientations of the

6 Scientific Programming

Gyros Accelerometers Magnetometers

Online calibration Offline calibration

Orientations estimation based QEKF

Constraint

Relative orientations

Absolute orientations

Kinematic parameters

Positions

Yes No

Kinematics of arm-hand

Figure 3 The diagram of the proposed method

first coordinate and q119895 is the quaternion of the absoluteorientations of the second coordinate

Meanwhile the kinematic models of the arm hand andfingers are considered to determine the orientations of eachsegment Human arm-hand motion can be supposed to bethe articulatedmotion of rigid body partsThese segments areupper arm (between the shoulder and elbow joints) forearm(between the elbow and wrist joints) hand (between thewrist joint and proximal joint) proximal finger (between theproximal joint and medial joint) medial finger (between themedial joint and distal joint) and distal finger (from the distaljoint) Every joint has its own local frame Shoulder can bemodeled as a ball joint with three DOFs and a fixed pointrepresenting the center of the shoulderMovements are repre-sented as the vector between the upper arm and body Elbowis the rotating hinge joint with two DOFs Wrist is modeledas a rotating hinge joint with two DOFs that is calculatedbetween the vector representing the hand and forearm Prox-imal joint is modeled as rotating hinge joint with two DOFsMedial joint and distal joint aremodeled as rotating joint withone DOF Thus the kinematic of this model consists of tenDOFs three in the shoulder joint two in the elbow joint twoin the wrist joint two in the proximal joint one in the medialjoint and one in the distal joint Hence the respondingconstraints are used to determine the orientations of eachsegment

34 Positions Estimation We assume the body keeps staticand the motion of arm and hand is formed by rotation of thejoints Hence the position of the fingertip p1198731198653 expressed in

the hand coordinate frame (see Figure 2) can be derived usingforward kinematics

[p11987311986531 ] = T1198731198601T11986011198602T1198602119867T1198671198651T11986511198652T11986521198653p11986531198652

= T1198731198653 [p119865311986521 ] (24)

where the transformation between two consecutive bodies isexpressed by T1198731198601 T11986011198602 T1198602119867 T1198671198651 T11986511198652 and T11986521198653

The total transformation T1198731198653 is given by the product ofeach consecutive contribution

T1198731198653 = [119877 (1199021198731198653) p119873119865301198793 1 ] (25)

where 119877(1199021198731198653) is the orientation of the distal phalanx withrespect to the body and p1198731198653 is the position of the distal frameexpressed in the global frame

35 Summary The gestures capture method is proposed inthe section The two procedures of the calibration are firstlyimplemented to improve the measurements of the sensorsThe inertial sensors are calibrated by the offline calibrationand the gyros are calibrated by the online calibration Thenthe orientations of the IMMUs are estimated by the QEKFAnd the kinematics of arm-hand is considered and the con-straints are integrated to determine the relative orientationsFinally the kinematic parameters are used to estimate thepositions A complete diagram of the method is depicted inFigure 3

Scientific Programming 7

minus20minus10

010

20

minus20minus10

010

20minus15

minus10

minus5

0

5

10

15z

(ms

2)

y (ms 2) x (ms2 )

Raw dataCalibrated

(a) The calibration results of accelerometers

minus1000minus500

0500

1000

minus1000minus500

0500

1000

minus1000

minus500

0

500

1000

Raw dataCalibrated

z(n

T)

y (nT) x (nT)

(b) The calibration results of magnetometers

Figure 4 The calibration results

Table 1 The calibration results

Scale Bias

Accelerometers G119886 = [[[[[

1 minus002 00 095 0120 0 092

]]]]]

b119886 = [[[[[

056087minus087

]]]]]

Magnetometers Gℎ = [[[[[

111 002 00 110 00 0 105

]]]]]

bℎ = [[[[[

571minus3954minus8298

]]]]]

4 Experiments and Results

The calibration is implemented to improve the sensors accu-racy and then the comparison experiments are testified toinvestigate the stability and accuracy of the orientations esti-mation The real-time gesture motion capture experimentsprove the validness of the proposed device

41 Calibration Results The calibration results of the triaxisaccelerometers and triaxis magnetometers are presented inthe section The calibration samples of IMMUs are collectedby rotating in various orientations Then the proposedmethod was used to determine the correctional calibrationparameters as G119886 Gℎ b119886 and bℎ The outputs of theaccelerometers and magnetometers are shown in Figure 4The blue sphere in Figure 4 is the raw date of sensors and thered sphere is the calibrated data of sensors The calibrationparameters are listed in Table 1

42 Evaluation Experiments After calibrating all the sensorsof the device the evaluation experiments are designed toassess the accuracy of the orientation estimation One of thestrings of the device is attached to the robotic arm ThreeIMMUs are set the same directions shown in Figure 5 Andthe robotic arm is rotated from the first state to the secondstate Then the estimated orientations of IMMUs compare

Table 2 The RMSE of the orientations

Static (∘) Dynamic (∘)Yaw Pitch Roll Yaw Pitch Roll

IMMU-1 004 001 002 243 005 102IMMU-2 011 001 002 168 051 103IMMU-3 010 002 002 177 002 007

with the true orientations of robotic arm The results of theorientations of three IMMUs are shown in Figure 6 Andthe results of the root mean square error (RMSE) of theorientations are listed in Table 2

As shown in Figure 6 during 5sim8 s the robotic armis dynamic And in static case it can be known that theresults have smaller variance and higher precision than theresults of the dynamic case Moreover the yaw angle has thelager variance compared with the roll and pitch It shouldbe caused by the motors of the robotic arm because themagnetometers are disturbed When the robotic arm is instatic the device can compute the accuracy orientations Thecomparison results proved the accuracy of the orientations ofthe IMMUs of the device Then the real-time gesture motioncapture experiments are implemented

43 Gesture Capture Experiments The IMMUsrsquo data is sam-pled collected and computed by the MCU and subsequentlytransmitted via Bluetooth to the external devices The MCUprocesses the raw data estimates the orientations of the eachunit encapsulates them into a packet and then sends thepacket to the PC by BluetoothThe baud rate for transmittingdata is 115200 bps The frequency is 50HZ By using thisdesign themotion capture can be demonstrated by the virtualmodel on the PC immediately The interface is written by CThe wearable system is shown as Figure 7

To further verify the effectiveness of proposed deviceupper arm is swung up and down The results of theorientations of the gestures are shown in Figure 8 And the

8 Scientific Programming

Robotic arm

IMMU-1

IMMU-2

IMMU-3

z1

x1

y1

(a) The first state

(b) The second state

Figure 5 The comparison experiments with robotic arm

positons of the fingertip of the right forefinger are shownin Figure 9 As shown in the figures the wearable devicecan determine the realistic movements of the arm andhand Meanwhile the accuracy of the results is assessed bythe statistics The root mean square error (RMSE) of theorientations is listed in Table 3 The RMSE of the positonsare listed in Table 4 Furthermore the wearable device isalso tested by ten healthy participants the real-time motiongestures capture experiments are implemented to prove thevalidity of the proposed system

44 Discussion In this paper a novel wearable device forgestures capturing based on magnetic and inertial measure-ment units is proposed As a practically useful applicationthe proposed device satisfies the requirements including theaccuracy computational efficiency and robustness First thecalibration method is designed to improve the accuracy of

0 5 10 15 20 25minus200

0

200

Time (s)

IMMU-1

YawPitch

Roll

YawPitch

Roll

YawPitch

Roll

IMMU-2

IMMU-3

Ang

el (∘

)

0 5 10 15 20 25minus200

0

200

Time (s)

Ang

el (∘

)0 5 10 15 20 25

minus200

0

200

Time (s)

Ang

el (∘

)

Figure 6 The results of orientations of the IMMUs

IMMU

Bluetooth

Data glovePC

Virtual modelTrajectory

Figure 7 The wearable system

Table 3 The RMSE of the orientations of the arm-hand

Angle (∘)Upper arm 107 110 044forearm 023 024Palm 019 018Proximal 011 026Medial 010Distal 008

Table 4 The RMSE of the positions of the fingertip

119909 (cm) 119910 (cm) 119911 (cm)Positions of the fingertip 345 135 234

the measurements of the sensors In particular the magneticeffects of the field are attenuated in advance Then the

Scientific Programming 9

0 10 20 30 40minus100

0

100Upper arm

t (s)0 10 20 30 40

minus1

0

1Forearm

t (s)

0 10 20 30 40minus1

0

1Palm

t (s)0 10 20 30 40

minus2

0

2Proximal

t (s)

0 10 20 30 40

0

05Medial

t (s)0 10 20 30 40

minus05minus05

0

05Distal

t (s)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Figure 8 The orientations of gestures

2040

6080 minus15 minus1 minus05 0 05 1

minus20

0

20

40

60

y (cm)x (cm)

z (c

m)

StartEnd

Figure 9 The positions of the fingertip of the right forefinger

orientations of gesture are divided into absolute orienta-tions and relative orientations The absolute orientations aredetermined by the QEKFs and the relative orientations areestimated by integrating the kinematics of the arm-handThe

positions are then easily computed The proposed method issimple and fast The algorithm is operated in the embeddedsystem at 50Hz The results of the experiments proved theadvantages of the proposed device

5 Conclusion

This paper presents the design implementation and exper-imental results of a wearable device for gestures captur-ing using inertial and magnetic sensor modules containingorthogonally mounted triads of accelerometers angular ratesensors and magnetometers Different from commercialmotion data gloves which usually use high-cost motion-sensing fibers to acquire hand motion data we adoptedthe low-cost inertial and magnetic sensor to reduce costMeanwhile the performance of the low-cost low-power andlight-weight IMMU is superior to some commercial IMMUFurthermore the novel device is proposed based on thirty-six IMMUs which cover the whole segments of the two armsand hands We designed the online and offline calibrationmethods to improve the accuracy of the units We deducedthe 3D arm and hand motion estimation algorithms thatintegrated the proposed kinematic models of the arm handand fingers and the attitude of gesture and positions offingertips can be determined For real-time performance and

10 Scientific Programming

convenience the interface with virtual model is designedPerformance evaluations verified that the proposed dataglove can accurately capture the motion of gestures Thesystem is developed in the way that all electronic componentscan be integrated and easy to wear This makes it moreconvenient and appealing for the user

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work has been supported by the National Science Foun-dation for Young Scientists of China (Grant no 61503212) andNational Natural Science Foundation of China (Grant nosU1613212 61327809 61210013 and 91420302)

References

[1] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) pp 25ndash28 Kitakyushu Japan August 2014

[2] J M Palacios C Sagues E Montijano and S LlorenteldquoHuman-computer interaction based on hand gestures usingRGB-D sensorsrdquo Sensors vol 13 no 9 pp 11842ndash11860 2013

[3] S S Rautaray and A Agrawal ldquoVision based hand gesturerecognition for human computer interaction a surveyrdquo Artifi-cial Intelligence Review vol 43 no 1 pp 1ndash54 2012

[4] E A Arkenbout J C F de Winter and P Breedveld ldquoRobusthandmotion tracking through data fusion of 5dt data glove andnimble VR kinect camera measurementsrdquo Sensors vol 15 no12 pp 31644ndash31671 2015

[5] D Regazzoni G De Vecchi and C Rizzi ldquoRGB cams vs RGB-D sensors low cost motion capture technologies performancesand limitationsrdquo Journal of Manufacturing Systems vol 33 no4 pp 719ndash728 2014

[6] Y Li X Chen X Zhang K Wang and Z J Wang ldquoA sign-component-based framework for Chinese sign language recog-nition using accelerometer and sEMG datardquo IEEE Transactionson Biomedical Engineering vol 59 no 10 pp 2695ndash2704 2012

[7] L Dipietro A M Sabatini and P Dario ldquoA survey of glove-based systems and their applicationsrdquo IEEE Transactions onSystems Man and Cybernetics Part C Applications and Reviewsvol 38 no 4 pp 461ndash482 2008

[8] B Takacs ldquoHow and why affordable virtual reality shapes thefuture of educationrdquoThe International Journal of Virtual Realityvol 7 no 1 pp 53ndash66 2008

[9] G Saggio ldquoA novel array of flex sensors for a goniometric gloverdquoSensors and Actuators A Physical vol 205 pp 119ndash125 2014

[10] R Gentner and J Classen ldquoDevelopment and evaluation of alow-cost sensor glove for assessment of human finger move-ments in neurophysiological settingsrdquo Journal of NeuroscienceMethods vol 178 no 1 pp 138ndash147 2009

[11] X L Wang and C L Yang ldquoConstructing gyro-free inertialmeasurement unit from dual accelerometers for gesture detec-tionrdquo Sensors amp Transducers vol 171 no 5 pp 134ndash140 2014

[12] W Chou B Fang L Ding X Ma and X Guo ldquoTwo-stepoptimal filter design for the low-cost attitude and headingreference systemsrdquo IET Science Measurement and Technologyvol 7 no 4 pp 240ndash248 2013

[13] D Roetenberg H J Luinge C T M Baten and P H VeltinkldquoCompensation of magnetic disturbances improves inertial andmagnetic sensing of human body segment orientationrdquo IEEETransactions on Neural Systems and Rehabilitation Engineeringvol 13 no 3 pp 395ndash405 2005

[14] J M Lambrecht and R F Kirsch ldquoMiniature low-powerinertial sensors promising technology for implantable motioncapture systemsrdquo IEEE Transactions on Neural Systems andRehabilitation Engineering vol 22 no 6 pp 1138ndash1147 2014

[15] R Xu S L Zhou and W J Li ldquoMEMS accelerometer basednonspecific-user hand gesture recognitionrdquo IEEE Sensors Jour-nal vol 12 no 5 pp 1166ndash1173 2012

[16] E Morganti L Angelini A Adami D Lalanne L Lorenzelliand E Mugellini ldquoA smart watch with embedded sensorsto recognize objects grasps and forearm gesturesrdquo ProcediaEngineering vol 41 pp 1169ndash1175 2012

[17] J-H Kim N DThang and T-S Kim ldquo3-D handmotion track-ing and gesture recognition using a data gloverdquo in Proceedingsof the IEEE International Symposium on Industrial Electronics(ISIE rsquo09) pp 1013ndash1018 IEEE Seoul South Korea July 2009

[18] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) vol 10 pp 25ndash28 Kitakyushu Japan August2014

[19] F Cavallo D Esposito E Rovini et al ldquoPreliminary evalu-ation of SensHand V1 in assessing motor skills performancein Parkinson diseaserdquo in Proceedings of the 2013 IEEE 13thInternational Conference onRehabilitation Robotics (ICORR rsquo13)pp 1ndash6 Seattle Wash USA June 2013

[20] H G Kortier V I Sluiter D Roetenberg and P H VeltinkldquoAssessment of hand kinematics using inertial and magneticsensorsrdquo Journal of NeuroEngineering and Rehabilitation vol 11no 1 article 70 2014

[21] Y Xu Y Wang Y Su and X Zhu ldquoResearch on the calibrationmethod of micro inertial measurement unit for engineeringapplicationrdquo Journal of Sensors vol 2016 Article ID 9108197 11pages 2016

[22] J Keighobadi ldquoFuzzy calibration of a magnetic compass forvehicular applicationsrdquoMechanical Systems and Signal Process-ing vol 25 no 6 pp 1973ndash1987 2011

[23] X Yun and E R Bachmann ldquoDesign implementation andexperimental results of a quaternion-based Kalman filter forhuman body motion trackingrdquo IEEE Transactions on Roboticsvol 22 no 6 pp 1216ndash1227 2006

[24] Xsens Technologies httpwwwxsenscom[25] T Hamel and R Mahony ldquoAttitude estimation on SO(3) based

on direct inertial measurementsrdquo in Proceedings of the IEEEInternational Conference on Robotics and Automation (ICRArsquo06) pp 2170ndash2175 Orlando Fla USA May 2006

[26] S Bonnabel PMartin and P Rouchon ldquoNon-linear symmetry-preserving observers on Lie groupsrdquo IEEE Transactions onAutomatic Control vol 54 no 7 pp 1709ndash1713 2009

[27] InvenSense MPU-9250 Nine-Axis MEMS MotionTrackingDevice 2015 httpwwwinvensensecom-productsmotion-tracking9-axismpu-9250

6 Scientific Programming

Gyros Accelerometers Magnetometers

Online calibration Offline calibration

Orientations estimation based QEKF

Constraint

Relative orientations

Absolute orientations

Kinematic parameters

Positions

Yes No

Kinematics of arm-hand

Figure 3 The diagram of the proposed method

first coordinate and q119895 is the quaternion of the absoluteorientations of the second coordinate

Meanwhile the kinematic models of the arm hand andfingers are considered to determine the orientations of eachsegment Human arm-hand motion can be supposed to bethe articulatedmotion of rigid body partsThese segments areupper arm (between the shoulder and elbow joints) forearm(between the elbow and wrist joints) hand (between thewrist joint and proximal joint) proximal finger (between theproximal joint and medial joint) medial finger (between themedial joint and distal joint) and distal finger (from the distaljoint) Every joint has its own local frame Shoulder can bemodeled as a ball joint with three DOFs and a fixed pointrepresenting the center of the shoulderMovements are repre-sented as the vector between the upper arm and body Elbowis the rotating hinge joint with two DOFs Wrist is modeledas a rotating hinge joint with two DOFs that is calculatedbetween the vector representing the hand and forearm Prox-imal joint is modeled as rotating hinge joint with two DOFsMedial joint and distal joint aremodeled as rotating joint withone DOF Thus the kinematic of this model consists of tenDOFs three in the shoulder joint two in the elbow joint twoin the wrist joint two in the proximal joint one in the medialjoint and one in the distal joint Hence the respondingconstraints are used to determine the orientations of eachsegment

34 Positions Estimation We assume the body keeps staticand the motion of arm and hand is formed by rotation of thejoints Hence the position of the fingertip p1198731198653 expressed in

the hand coordinate frame (see Figure 2) can be derived usingforward kinematics

[p11987311986531 ] = T1198731198601T11986011198602T1198602119867T1198671198651T11986511198652T11986521198653p11986531198652

= T1198731198653 [p119865311986521 ] (24)

where the transformation between two consecutive bodies isexpressed by T1198731198601 T11986011198602 T1198602119867 T1198671198651 T11986511198652 and T11986521198653

The total transformation T1198731198653 is given by the product ofeach consecutive contribution

T1198731198653 = [119877 (1199021198731198653) p119873119865301198793 1 ] (25)

where 119877(1199021198731198653) is the orientation of the distal phalanx withrespect to the body and p1198731198653 is the position of the distal frameexpressed in the global frame

35 Summary The gestures capture method is proposed inthe section The two procedures of the calibration are firstlyimplemented to improve the measurements of the sensorsThe inertial sensors are calibrated by the offline calibrationand the gyros are calibrated by the online calibration Thenthe orientations of the IMMUs are estimated by the QEKFAnd the kinematics of arm-hand is considered and the con-straints are integrated to determine the relative orientationsFinally the kinematic parameters are used to estimate thepositions A complete diagram of the method is depicted inFigure 3

Scientific Programming 7

minus20minus10

010

20

minus20minus10

010

20minus15

minus10

minus5

0

5

10

15z

(ms

2)

y (ms 2) x (ms2 )

Raw dataCalibrated

(a) The calibration results of accelerometers

minus1000minus500

0500

1000

minus1000minus500

0500

1000

minus1000

minus500

0

500

1000

Raw dataCalibrated

z(n

T)

y (nT) x (nT)

(b) The calibration results of magnetometers

Figure 4 The calibration results

Table 1 The calibration results

Scale Bias

Accelerometers G119886 = [[[[[

1 minus002 00 095 0120 0 092

]]]]]

b119886 = [[[[[

056087minus087

]]]]]

Magnetometers Gℎ = [[[[[

111 002 00 110 00 0 105

]]]]]

bℎ = [[[[[

571minus3954minus8298

]]]]]

4 Experiments and Results

The calibration is implemented to improve the sensors accu-racy and then the comparison experiments are testified toinvestigate the stability and accuracy of the orientations esti-mation The real-time gesture motion capture experimentsprove the validness of the proposed device

41 Calibration Results The calibration results of the triaxisaccelerometers and triaxis magnetometers are presented inthe section The calibration samples of IMMUs are collectedby rotating in various orientations Then the proposedmethod was used to determine the correctional calibrationparameters as G119886 Gℎ b119886 and bℎ The outputs of theaccelerometers and magnetometers are shown in Figure 4The blue sphere in Figure 4 is the raw date of sensors and thered sphere is the calibrated data of sensors The calibrationparameters are listed in Table 1

42 Evaluation Experiments After calibrating all the sensorsof the device the evaluation experiments are designed toassess the accuracy of the orientation estimation One of thestrings of the device is attached to the robotic arm ThreeIMMUs are set the same directions shown in Figure 5 Andthe robotic arm is rotated from the first state to the secondstate Then the estimated orientations of IMMUs compare

Table 2 The RMSE of the orientations

Static (∘) Dynamic (∘)Yaw Pitch Roll Yaw Pitch Roll

IMMU-1 004 001 002 243 005 102IMMU-2 011 001 002 168 051 103IMMU-3 010 002 002 177 002 007

with the true orientations of robotic arm The results of theorientations of three IMMUs are shown in Figure 6 Andthe results of the root mean square error (RMSE) of theorientations are listed in Table 2

As shown in Figure 6 during 5sim8 s the robotic armis dynamic And in static case it can be known that theresults have smaller variance and higher precision than theresults of the dynamic case Moreover the yaw angle has thelager variance compared with the roll and pitch It shouldbe caused by the motors of the robotic arm because themagnetometers are disturbed When the robotic arm is instatic the device can compute the accuracy orientations Thecomparison results proved the accuracy of the orientations ofthe IMMUs of the device Then the real-time gesture motioncapture experiments are implemented

43 Gesture Capture Experiments The IMMUsrsquo data is sam-pled collected and computed by the MCU and subsequentlytransmitted via Bluetooth to the external devices The MCUprocesses the raw data estimates the orientations of the eachunit encapsulates them into a packet and then sends thepacket to the PC by BluetoothThe baud rate for transmittingdata is 115200 bps The frequency is 50HZ By using thisdesign themotion capture can be demonstrated by the virtualmodel on the PC immediately The interface is written by CThe wearable system is shown as Figure 7

To further verify the effectiveness of proposed deviceupper arm is swung up and down The results of theorientations of the gestures are shown in Figure 8 And the

8 Scientific Programming

Robotic arm

IMMU-1

IMMU-2

IMMU-3

z1

x1

y1

(a) The first state

(b) The second state

Figure 5 The comparison experiments with robotic arm

positons of the fingertip of the right forefinger are shownin Figure 9 As shown in the figures the wearable devicecan determine the realistic movements of the arm andhand Meanwhile the accuracy of the results is assessed bythe statistics The root mean square error (RMSE) of theorientations is listed in Table 3 The RMSE of the positonsare listed in Table 4 Furthermore the wearable device isalso tested by ten healthy participants the real-time motiongestures capture experiments are implemented to prove thevalidity of the proposed system

44 Discussion In this paper a novel wearable device forgestures capturing based on magnetic and inertial measure-ment units is proposed As a practically useful applicationthe proposed device satisfies the requirements including theaccuracy computational efficiency and robustness First thecalibration method is designed to improve the accuracy of

0 5 10 15 20 25minus200

0

200

Time (s)

IMMU-1

YawPitch

Roll

YawPitch

Roll

YawPitch

Roll

IMMU-2

IMMU-3

Ang

el (∘

)

0 5 10 15 20 25minus200

0

200

Time (s)

Ang

el (∘

)0 5 10 15 20 25

minus200

0

200

Time (s)

Ang

el (∘

)

Figure 6 The results of orientations of the IMMUs

IMMU

Bluetooth

Data glovePC

Virtual modelTrajectory

Figure 7 The wearable system

Table 3 The RMSE of the orientations of the arm-hand

Angle (∘)Upper arm 107 110 044forearm 023 024Palm 019 018Proximal 011 026Medial 010Distal 008

Table 4 The RMSE of the positions of the fingertip

119909 (cm) 119910 (cm) 119911 (cm)Positions of the fingertip 345 135 234

the measurements of the sensors In particular the magneticeffects of the field are attenuated in advance Then the

Scientific Programming 9

0 10 20 30 40minus100

0

100Upper arm

t (s)0 10 20 30 40

minus1

0

1Forearm

t (s)

0 10 20 30 40minus1

0

1Palm

t (s)0 10 20 30 40

minus2

0

2Proximal

t (s)

0 10 20 30 40

0

05Medial

t (s)0 10 20 30 40

minus05minus05

0

05Distal

t (s)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Figure 8 The orientations of gestures

2040

6080 minus15 minus1 minus05 0 05 1

minus20

0

20

40

60

y (cm)x (cm)

z (c

m)

StartEnd

Figure 9 The positions of the fingertip of the right forefinger

orientations of gesture are divided into absolute orienta-tions and relative orientations The absolute orientations aredetermined by the QEKFs and the relative orientations areestimated by integrating the kinematics of the arm-handThe

positions are then easily computed The proposed method issimple and fast The algorithm is operated in the embeddedsystem at 50Hz The results of the experiments proved theadvantages of the proposed device

5 Conclusion

This paper presents the design implementation and exper-imental results of a wearable device for gestures captur-ing using inertial and magnetic sensor modules containingorthogonally mounted triads of accelerometers angular ratesensors and magnetometers Different from commercialmotion data gloves which usually use high-cost motion-sensing fibers to acquire hand motion data we adoptedthe low-cost inertial and magnetic sensor to reduce costMeanwhile the performance of the low-cost low-power andlight-weight IMMU is superior to some commercial IMMUFurthermore the novel device is proposed based on thirty-six IMMUs which cover the whole segments of the two armsand hands We designed the online and offline calibrationmethods to improve the accuracy of the units We deducedthe 3D arm and hand motion estimation algorithms thatintegrated the proposed kinematic models of the arm handand fingers and the attitude of gesture and positions offingertips can be determined For real-time performance and

10 Scientific Programming

convenience the interface with virtual model is designedPerformance evaluations verified that the proposed dataglove can accurately capture the motion of gestures Thesystem is developed in the way that all electronic componentscan be integrated and easy to wear This makes it moreconvenient and appealing for the user

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work has been supported by the National Science Foun-dation for Young Scientists of China (Grant no 61503212) andNational Natural Science Foundation of China (Grant nosU1613212 61327809 61210013 and 91420302)

References

[1] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) pp 25ndash28 Kitakyushu Japan August 2014

[2] J M Palacios C Sagues E Montijano and S LlorenteldquoHuman-computer interaction based on hand gestures usingRGB-D sensorsrdquo Sensors vol 13 no 9 pp 11842ndash11860 2013

[3] S S Rautaray and A Agrawal ldquoVision based hand gesturerecognition for human computer interaction a surveyrdquo Artifi-cial Intelligence Review vol 43 no 1 pp 1ndash54 2012

[4] E A Arkenbout J C F de Winter and P Breedveld ldquoRobusthandmotion tracking through data fusion of 5dt data glove andnimble VR kinect camera measurementsrdquo Sensors vol 15 no12 pp 31644ndash31671 2015

[5] D Regazzoni G De Vecchi and C Rizzi ldquoRGB cams vs RGB-D sensors low cost motion capture technologies performancesand limitationsrdquo Journal of Manufacturing Systems vol 33 no4 pp 719ndash728 2014

[6] Y Li X Chen X Zhang K Wang and Z J Wang ldquoA sign-component-based framework for Chinese sign language recog-nition using accelerometer and sEMG datardquo IEEE Transactionson Biomedical Engineering vol 59 no 10 pp 2695ndash2704 2012

[7] L Dipietro A M Sabatini and P Dario ldquoA survey of glove-based systems and their applicationsrdquo IEEE Transactions onSystems Man and Cybernetics Part C Applications and Reviewsvol 38 no 4 pp 461ndash482 2008

[8] B Takacs ldquoHow and why affordable virtual reality shapes thefuture of educationrdquoThe International Journal of Virtual Realityvol 7 no 1 pp 53ndash66 2008

[9] G Saggio ldquoA novel array of flex sensors for a goniometric gloverdquoSensors and Actuators A Physical vol 205 pp 119ndash125 2014

[10] R Gentner and J Classen ldquoDevelopment and evaluation of alow-cost sensor glove for assessment of human finger move-ments in neurophysiological settingsrdquo Journal of NeuroscienceMethods vol 178 no 1 pp 138ndash147 2009

[11] X L Wang and C L Yang ldquoConstructing gyro-free inertialmeasurement unit from dual accelerometers for gesture detec-tionrdquo Sensors amp Transducers vol 171 no 5 pp 134ndash140 2014

[12] W Chou B Fang L Ding X Ma and X Guo ldquoTwo-stepoptimal filter design for the low-cost attitude and headingreference systemsrdquo IET Science Measurement and Technologyvol 7 no 4 pp 240ndash248 2013

[13] D Roetenberg H J Luinge C T M Baten and P H VeltinkldquoCompensation of magnetic disturbances improves inertial andmagnetic sensing of human body segment orientationrdquo IEEETransactions on Neural Systems and Rehabilitation Engineeringvol 13 no 3 pp 395ndash405 2005

[14] J M Lambrecht and R F Kirsch ldquoMiniature low-powerinertial sensors promising technology for implantable motioncapture systemsrdquo IEEE Transactions on Neural Systems andRehabilitation Engineering vol 22 no 6 pp 1138ndash1147 2014

[15] R Xu S L Zhou and W J Li ldquoMEMS accelerometer basednonspecific-user hand gesture recognitionrdquo IEEE Sensors Jour-nal vol 12 no 5 pp 1166ndash1173 2012

[16] E Morganti L Angelini A Adami D Lalanne L Lorenzelliand E Mugellini ldquoA smart watch with embedded sensorsto recognize objects grasps and forearm gesturesrdquo ProcediaEngineering vol 41 pp 1169ndash1175 2012

[17] J-H Kim N DThang and T-S Kim ldquo3-D handmotion track-ing and gesture recognition using a data gloverdquo in Proceedingsof the IEEE International Symposium on Industrial Electronics(ISIE rsquo09) pp 1013ndash1018 IEEE Seoul South Korea July 2009

[18] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) vol 10 pp 25ndash28 Kitakyushu Japan August2014

[19] F Cavallo D Esposito E Rovini et al ldquoPreliminary evalu-ation of SensHand V1 in assessing motor skills performancein Parkinson diseaserdquo in Proceedings of the 2013 IEEE 13thInternational Conference onRehabilitation Robotics (ICORR rsquo13)pp 1ndash6 Seattle Wash USA June 2013

[20] H G Kortier V I Sluiter D Roetenberg and P H VeltinkldquoAssessment of hand kinematics using inertial and magneticsensorsrdquo Journal of NeuroEngineering and Rehabilitation vol 11no 1 article 70 2014

[21] Y Xu Y Wang Y Su and X Zhu ldquoResearch on the calibrationmethod of micro inertial measurement unit for engineeringapplicationrdquo Journal of Sensors vol 2016 Article ID 9108197 11pages 2016

[22] J Keighobadi ldquoFuzzy calibration of a magnetic compass forvehicular applicationsrdquoMechanical Systems and Signal Process-ing vol 25 no 6 pp 1973ndash1987 2011

[23] X Yun and E R Bachmann ldquoDesign implementation andexperimental results of a quaternion-based Kalman filter forhuman body motion trackingrdquo IEEE Transactions on Roboticsvol 22 no 6 pp 1216ndash1227 2006

[24] Xsens Technologies httpwwwxsenscom[25] T Hamel and R Mahony ldquoAttitude estimation on SO(3) based

on direct inertial measurementsrdquo in Proceedings of the IEEEInternational Conference on Robotics and Automation (ICRArsquo06) pp 2170ndash2175 Orlando Fla USA May 2006

[26] S Bonnabel PMartin and P Rouchon ldquoNon-linear symmetry-preserving observers on Lie groupsrdquo IEEE Transactions onAutomatic Control vol 54 no 7 pp 1709ndash1713 2009

[27] InvenSense MPU-9250 Nine-Axis MEMS MotionTrackingDevice 2015 httpwwwinvensensecom-productsmotion-tracking9-axismpu-9250

Scientific Programming 7

minus20minus10

010

20

minus20minus10

010

20minus15

minus10

minus5

0

5

10

15z

(ms

2)

y (ms 2) x (ms2 )

Raw dataCalibrated

(a) The calibration results of accelerometers

minus1000minus500

0500

1000

minus1000minus500

0500

1000

minus1000

minus500

0

500

1000

Raw dataCalibrated

z(n

T)

y (nT) x (nT)

(b) The calibration results of magnetometers

Figure 4 The calibration results

Table 1 The calibration results

Scale Bias

Accelerometers G119886 = [[[[[

1 minus002 00 095 0120 0 092

]]]]]

b119886 = [[[[[

056087minus087

]]]]]

Magnetometers Gℎ = [[[[[

111 002 00 110 00 0 105

]]]]]

bℎ = [[[[[

571minus3954minus8298

]]]]]

4 Experiments and Results

The calibration is implemented to improve the sensors accu-racy and then the comparison experiments are testified toinvestigate the stability and accuracy of the orientations esti-mation The real-time gesture motion capture experimentsprove the validness of the proposed device

41 Calibration Results The calibration results of the triaxisaccelerometers and triaxis magnetometers are presented inthe section The calibration samples of IMMUs are collectedby rotating in various orientations Then the proposedmethod was used to determine the correctional calibrationparameters as G119886 Gℎ b119886 and bℎ The outputs of theaccelerometers and magnetometers are shown in Figure 4The blue sphere in Figure 4 is the raw date of sensors and thered sphere is the calibrated data of sensors The calibrationparameters are listed in Table 1

42 Evaluation Experiments After calibrating all the sensorsof the device the evaluation experiments are designed toassess the accuracy of the orientation estimation One of thestrings of the device is attached to the robotic arm ThreeIMMUs are set the same directions shown in Figure 5 Andthe robotic arm is rotated from the first state to the secondstate Then the estimated orientations of IMMUs compare

Table 2 The RMSE of the orientations

Static (∘) Dynamic (∘)Yaw Pitch Roll Yaw Pitch Roll

IMMU-1 004 001 002 243 005 102IMMU-2 011 001 002 168 051 103IMMU-3 010 002 002 177 002 007

with the true orientations of robotic arm The results of theorientations of three IMMUs are shown in Figure 6 Andthe results of the root mean square error (RMSE) of theorientations are listed in Table 2

As shown in Figure 6 during 5sim8 s the robotic armis dynamic And in static case it can be known that theresults have smaller variance and higher precision than theresults of the dynamic case Moreover the yaw angle has thelager variance compared with the roll and pitch It shouldbe caused by the motors of the robotic arm because themagnetometers are disturbed When the robotic arm is instatic the device can compute the accuracy orientations Thecomparison results proved the accuracy of the orientations ofthe IMMUs of the device Then the real-time gesture motioncapture experiments are implemented

43 Gesture Capture Experiments The IMMUsrsquo data is sam-pled collected and computed by the MCU and subsequentlytransmitted via Bluetooth to the external devices The MCUprocesses the raw data estimates the orientations of the eachunit encapsulates them into a packet and then sends thepacket to the PC by BluetoothThe baud rate for transmittingdata is 115200 bps The frequency is 50HZ By using thisdesign themotion capture can be demonstrated by the virtualmodel on the PC immediately The interface is written by CThe wearable system is shown as Figure 7

To further verify the effectiveness of proposed deviceupper arm is swung up and down The results of theorientations of the gestures are shown in Figure 8 And the

8 Scientific Programming

Robotic arm

IMMU-1

IMMU-2

IMMU-3

z1

x1

y1

(a) The first state

(b) The second state

Figure 5 The comparison experiments with robotic arm

positons of the fingertip of the right forefinger are shownin Figure 9 As shown in the figures the wearable devicecan determine the realistic movements of the arm andhand Meanwhile the accuracy of the results is assessed bythe statistics The root mean square error (RMSE) of theorientations is listed in Table 3 The RMSE of the positonsare listed in Table 4 Furthermore the wearable device isalso tested by ten healthy participants the real-time motiongestures capture experiments are implemented to prove thevalidity of the proposed system

44 Discussion In this paper a novel wearable device forgestures capturing based on magnetic and inertial measure-ment units is proposed As a practically useful applicationthe proposed device satisfies the requirements including theaccuracy computational efficiency and robustness First thecalibration method is designed to improve the accuracy of

0 5 10 15 20 25minus200

0

200

Time (s)

IMMU-1

YawPitch

Roll

YawPitch

Roll

YawPitch

Roll

IMMU-2

IMMU-3

Ang

el (∘

)

0 5 10 15 20 25minus200

0

200

Time (s)

Ang

el (∘

)0 5 10 15 20 25

minus200

0

200

Time (s)

Ang

el (∘

)

Figure 6 The results of orientations of the IMMUs

IMMU

Bluetooth

Data glovePC

Virtual modelTrajectory

Figure 7 The wearable system

Table 3 The RMSE of the orientations of the arm-hand

Angle (∘)Upper arm 107 110 044forearm 023 024Palm 019 018Proximal 011 026Medial 010Distal 008

Table 4 The RMSE of the positions of the fingertip

119909 (cm) 119910 (cm) 119911 (cm)Positions of the fingertip 345 135 234

the measurements of the sensors In particular the magneticeffects of the field are attenuated in advance Then the

Scientific Programming 9

0 10 20 30 40minus100

0

100Upper arm

t (s)0 10 20 30 40

minus1

0

1Forearm

t (s)

0 10 20 30 40minus1

0

1Palm

t (s)0 10 20 30 40

minus2

0

2Proximal

t (s)

0 10 20 30 40

0

05Medial

t (s)0 10 20 30 40

minus05minus05

0

05Distal

t (s)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Figure 8 The orientations of gestures

2040

6080 minus15 minus1 minus05 0 05 1

minus20

0

20

40

60

y (cm)x (cm)

z (c

m)

StartEnd

Figure 9 The positions of the fingertip of the right forefinger

orientations of gesture are divided into absolute orienta-tions and relative orientations The absolute orientations aredetermined by the QEKFs and the relative orientations areestimated by integrating the kinematics of the arm-handThe

positions are then easily computed The proposed method issimple and fast The algorithm is operated in the embeddedsystem at 50Hz The results of the experiments proved theadvantages of the proposed device

5 Conclusion

This paper presents the design implementation and exper-imental results of a wearable device for gestures captur-ing using inertial and magnetic sensor modules containingorthogonally mounted triads of accelerometers angular ratesensors and magnetometers Different from commercialmotion data gloves which usually use high-cost motion-sensing fibers to acquire hand motion data we adoptedthe low-cost inertial and magnetic sensor to reduce costMeanwhile the performance of the low-cost low-power andlight-weight IMMU is superior to some commercial IMMUFurthermore the novel device is proposed based on thirty-six IMMUs which cover the whole segments of the two armsand hands We designed the online and offline calibrationmethods to improve the accuracy of the units We deducedthe 3D arm and hand motion estimation algorithms thatintegrated the proposed kinematic models of the arm handand fingers and the attitude of gesture and positions offingertips can be determined For real-time performance and

10 Scientific Programming

convenience the interface with virtual model is designedPerformance evaluations verified that the proposed dataglove can accurately capture the motion of gestures Thesystem is developed in the way that all electronic componentscan be integrated and easy to wear This makes it moreconvenient and appealing for the user

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work has been supported by the National Science Foun-dation for Young Scientists of China (Grant no 61503212) andNational Natural Science Foundation of China (Grant nosU1613212 61327809 61210013 and 91420302)

References

[1] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) pp 25ndash28 Kitakyushu Japan August 2014

[2] J M Palacios C Sagues E Montijano and S LlorenteldquoHuman-computer interaction based on hand gestures usingRGB-D sensorsrdquo Sensors vol 13 no 9 pp 11842ndash11860 2013

[3] S S Rautaray and A Agrawal ldquoVision based hand gesturerecognition for human computer interaction a surveyrdquo Artifi-cial Intelligence Review vol 43 no 1 pp 1ndash54 2012

[4] E A Arkenbout J C F de Winter and P Breedveld ldquoRobusthandmotion tracking through data fusion of 5dt data glove andnimble VR kinect camera measurementsrdquo Sensors vol 15 no12 pp 31644ndash31671 2015

[5] D Regazzoni G De Vecchi and C Rizzi ldquoRGB cams vs RGB-D sensors low cost motion capture technologies performancesand limitationsrdquo Journal of Manufacturing Systems vol 33 no4 pp 719ndash728 2014

[6] Y Li X Chen X Zhang K Wang and Z J Wang ldquoA sign-component-based framework for Chinese sign language recog-nition using accelerometer and sEMG datardquo IEEE Transactionson Biomedical Engineering vol 59 no 10 pp 2695ndash2704 2012

[7] L Dipietro A M Sabatini and P Dario ldquoA survey of glove-based systems and their applicationsrdquo IEEE Transactions onSystems Man and Cybernetics Part C Applications and Reviewsvol 38 no 4 pp 461ndash482 2008

[8] B Takacs ldquoHow and why affordable virtual reality shapes thefuture of educationrdquoThe International Journal of Virtual Realityvol 7 no 1 pp 53ndash66 2008

[9] G Saggio ldquoA novel array of flex sensors for a goniometric gloverdquoSensors and Actuators A Physical vol 205 pp 119ndash125 2014

[10] R Gentner and J Classen ldquoDevelopment and evaluation of alow-cost sensor glove for assessment of human finger move-ments in neurophysiological settingsrdquo Journal of NeuroscienceMethods vol 178 no 1 pp 138ndash147 2009

[11] X L Wang and C L Yang ldquoConstructing gyro-free inertialmeasurement unit from dual accelerometers for gesture detec-tionrdquo Sensors amp Transducers vol 171 no 5 pp 134ndash140 2014

[12] W Chou B Fang L Ding X Ma and X Guo ldquoTwo-stepoptimal filter design for the low-cost attitude and headingreference systemsrdquo IET Science Measurement and Technologyvol 7 no 4 pp 240ndash248 2013

[13] D Roetenberg H J Luinge C T M Baten and P H VeltinkldquoCompensation of magnetic disturbances improves inertial andmagnetic sensing of human body segment orientationrdquo IEEETransactions on Neural Systems and Rehabilitation Engineeringvol 13 no 3 pp 395ndash405 2005

[14] J M Lambrecht and R F Kirsch ldquoMiniature low-powerinertial sensors promising technology for implantable motioncapture systemsrdquo IEEE Transactions on Neural Systems andRehabilitation Engineering vol 22 no 6 pp 1138ndash1147 2014

[15] R Xu S L Zhou and W J Li ldquoMEMS accelerometer basednonspecific-user hand gesture recognitionrdquo IEEE Sensors Jour-nal vol 12 no 5 pp 1166ndash1173 2012

[16] E Morganti L Angelini A Adami D Lalanne L Lorenzelliand E Mugellini ldquoA smart watch with embedded sensorsto recognize objects grasps and forearm gesturesrdquo ProcediaEngineering vol 41 pp 1169ndash1175 2012

[17] J-H Kim N DThang and T-S Kim ldquo3-D handmotion track-ing and gesture recognition using a data gloverdquo in Proceedingsof the IEEE International Symposium on Industrial Electronics(ISIE rsquo09) pp 1013ndash1018 IEEE Seoul South Korea July 2009

[18] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) vol 10 pp 25ndash28 Kitakyushu Japan August2014

[19] F Cavallo D Esposito E Rovini et al ldquoPreliminary evalu-ation of SensHand V1 in assessing motor skills performancein Parkinson diseaserdquo in Proceedings of the 2013 IEEE 13thInternational Conference onRehabilitation Robotics (ICORR rsquo13)pp 1ndash6 Seattle Wash USA June 2013

[20] H G Kortier V I Sluiter D Roetenberg and P H VeltinkldquoAssessment of hand kinematics using inertial and magneticsensorsrdquo Journal of NeuroEngineering and Rehabilitation vol 11no 1 article 70 2014

[21] Y Xu Y Wang Y Su and X Zhu ldquoResearch on the calibrationmethod of micro inertial measurement unit for engineeringapplicationrdquo Journal of Sensors vol 2016 Article ID 9108197 11pages 2016

[22] J Keighobadi ldquoFuzzy calibration of a magnetic compass forvehicular applicationsrdquoMechanical Systems and Signal Process-ing vol 25 no 6 pp 1973ndash1987 2011

[23] X Yun and E R Bachmann ldquoDesign implementation andexperimental results of a quaternion-based Kalman filter forhuman body motion trackingrdquo IEEE Transactions on Roboticsvol 22 no 6 pp 1216ndash1227 2006

[24] Xsens Technologies httpwwwxsenscom[25] T Hamel and R Mahony ldquoAttitude estimation on SO(3) based

on direct inertial measurementsrdquo in Proceedings of the IEEEInternational Conference on Robotics and Automation (ICRArsquo06) pp 2170ndash2175 Orlando Fla USA May 2006

[26] S Bonnabel PMartin and P Rouchon ldquoNon-linear symmetry-preserving observers on Lie groupsrdquo IEEE Transactions onAutomatic Control vol 54 no 7 pp 1709ndash1713 2009

[27] InvenSense MPU-9250 Nine-Axis MEMS MotionTrackingDevice 2015 httpwwwinvensensecom-productsmotion-tracking9-axismpu-9250

8 Scientific Programming

Robotic arm

IMMU-1

IMMU-2

IMMU-3

z1

x1

y1

(a) The first state

(b) The second state

Figure 5 The comparison experiments with robotic arm

positons of the fingertip of the right forefinger are shownin Figure 9 As shown in the figures the wearable devicecan determine the realistic movements of the arm andhand Meanwhile the accuracy of the results is assessed bythe statistics The root mean square error (RMSE) of theorientations is listed in Table 3 The RMSE of the positonsare listed in Table 4 Furthermore the wearable device isalso tested by ten healthy participants the real-time motiongestures capture experiments are implemented to prove thevalidity of the proposed system

44 Discussion In this paper a novel wearable device forgestures capturing based on magnetic and inertial measure-ment units is proposed As a practically useful applicationthe proposed device satisfies the requirements including theaccuracy computational efficiency and robustness First thecalibration method is designed to improve the accuracy of

0 5 10 15 20 25minus200

0

200

Time (s)

IMMU-1

YawPitch

Roll

YawPitch

Roll

YawPitch

Roll

IMMU-2

IMMU-3

Ang

el (∘

)

0 5 10 15 20 25minus200

0

200

Time (s)

Ang

el (∘

)0 5 10 15 20 25

minus200

0

200

Time (s)

Ang

el (∘

)

Figure 6 The results of orientations of the IMMUs

IMMU

Bluetooth

Data glovePC

Virtual modelTrajectory

Figure 7 The wearable system

Table 3 The RMSE of the orientations of the arm-hand

Angle (∘)Upper arm 107 110 044forearm 023 024Palm 019 018Proximal 011 026Medial 010Distal 008

Table 4 The RMSE of the positions of the fingertip

119909 (cm) 119910 (cm) 119911 (cm)Positions of the fingertip 345 135 234

the measurements of the sensors In particular the magneticeffects of the field are attenuated in advance Then the

Scientific Programming 9

0 10 20 30 40minus100

0

100Upper arm

t (s)0 10 20 30 40

minus1

0

1Forearm

t (s)

0 10 20 30 40minus1

0

1Palm

t (s)0 10 20 30 40

minus2

0

2Proximal

t (s)

0 10 20 30 40

0

05Medial

t (s)0 10 20 30 40

minus05minus05

0

05Distal

t (s)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Figure 8 The orientations of gestures

2040

6080 minus15 minus1 minus05 0 05 1

minus20

0

20

40

60

y (cm)x (cm)

z (c

m)

StartEnd

Figure 9 The positions of the fingertip of the right forefinger

orientations of gesture are divided into absolute orienta-tions and relative orientations The absolute orientations aredetermined by the QEKFs and the relative orientations areestimated by integrating the kinematics of the arm-handThe

positions are then easily computed The proposed method issimple and fast The algorithm is operated in the embeddedsystem at 50Hz The results of the experiments proved theadvantages of the proposed device

5 Conclusion

This paper presents the design implementation and exper-imental results of a wearable device for gestures captur-ing using inertial and magnetic sensor modules containingorthogonally mounted triads of accelerometers angular ratesensors and magnetometers Different from commercialmotion data gloves which usually use high-cost motion-sensing fibers to acquire hand motion data we adoptedthe low-cost inertial and magnetic sensor to reduce costMeanwhile the performance of the low-cost low-power andlight-weight IMMU is superior to some commercial IMMUFurthermore the novel device is proposed based on thirty-six IMMUs which cover the whole segments of the two armsand hands We designed the online and offline calibrationmethods to improve the accuracy of the units We deducedthe 3D arm and hand motion estimation algorithms thatintegrated the proposed kinematic models of the arm handand fingers and the attitude of gesture and positions offingertips can be determined For real-time performance and

10 Scientific Programming

convenience the interface with virtual model is designedPerformance evaluations verified that the proposed dataglove can accurately capture the motion of gestures Thesystem is developed in the way that all electronic componentscan be integrated and easy to wear This makes it moreconvenient and appealing for the user

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work has been supported by the National Science Foun-dation for Young Scientists of China (Grant no 61503212) andNational Natural Science Foundation of China (Grant nosU1613212 61327809 61210013 and 91420302)

References

[1] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) pp 25ndash28 Kitakyushu Japan August 2014

[2] J M Palacios C Sagues E Montijano and S LlorenteldquoHuman-computer interaction based on hand gestures usingRGB-D sensorsrdquo Sensors vol 13 no 9 pp 11842ndash11860 2013

[3] S S Rautaray and A Agrawal ldquoVision based hand gesturerecognition for human computer interaction a surveyrdquo Artifi-cial Intelligence Review vol 43 no 1 pp 1ndash54 2012

[4] E A Arkenbout J C F de Winter and P Breedveld ldquoRobusthandmotion tracking through data fusion of 5dt data glove andnimble VR kinect camera measurementsrdquo Sensors vol 15 no12 pp 31644ndash31671 2015

[5] D Regazzoni G De Vecchi and C Rizzi ldquoRGB cams vs RGB-D sensors low cost motion capture technologies performancesand limitationsrdquo Journal of Manufacturing Systems vol 33 no4 pp 719ndash728 2014

[6] Y Li X Chen X Zhang K Wang and Z J Wang ldquoA sign-component-based framework for Chinese sign language recog-nition using accelerometer and sEMG datardquo IEEE Transactionson Biomedical Engineering vol 59 no 10 pp 2695ndash2704 2012

[7] L Dipietro A M Sabatini and P Dario ldquoA survey of glove-based systems and their applicationsrdquo IEEE Transactions onSystems Man and Cybernetics Part C Applications and Reviewsvol 38 no 4 pp 461ndash482 2008

[8] B Takacs ldquoHow and why affordable virtual reality shapes thefuture of educationrdquoThe International Journal of Virtual Realityvol 7 no 1 pp 53ndash66 2008

[9] G Saggio ldquoA novel array of flex sensors for a goniometric gloverdquoSensors and Actuators A Physical vol 205 pp 119ndash125 2014

[10] R Gentner and J Classen ldquoDevelopment and evaluation of alow-cost sensor glove for assessment of human finger move-ments in neurophysiological settingsrdquo Journal of NeuroscienceMethods vol 178 no 1 pp 138ndash147 2009

[11] X L Wang and C L Yang ldquoConstructing gyro-free inertialmeasurement unit from dual accelerometers for gesture detec-tionrdquo Sensors amp Transducers vol 171 no 5 pp 134ndash140 2014

[12] W Chou B Fang L Ding X Ma and X Guo ldquoTwo-stepoptimal filter design for the low-cost attitude and headingreference systemsrdquo IET Science Measurement and Technologyvol 7 no 4 pp 240ndash248 2013

[13] D Roetenberg H J Luinge C T M Baten and P H VeltinkldquoCompensation of magnetic disturbances improves inertial andmagnetic sensing of human body segment orientationrdquo IEEETransactions on Neural Systems and Rehabilitation Engineeringvol 13 no 3 pp 395ndash405 2005

[14] J M Lambrecht and R F Kirsch ldquoMiniature low-powerinertial sensors promising technology for implantable motioncapture systemsrdquo IEEE Transactions on Neural Systems andRehabilitation Engineering vol 22 no 6 pp 1138ndash1147 2014

[15] R Xu S L Zhou and W J Li ldquoMEMS accelerometer basednonspecific-user hand gesture recognitionrdquo IEEE Sensors Jour-nal vol 12 no 5 pp 1166ndash1173 2012

[16] E Morganti L Angelini A Adami D Lalanne L Lorenzelliand E Mugellini ldquoA smart watch with embedded sensorsto recognize objects grasps and forearm gesturesrdquo ProcediaEngineering vol 41 pp 1169ndash1175 2012

[17] J-H Kim N DThang and T-S Kim ldquo3-D handmotion track-ing and gesture recognition using a data gloverdquo in Proceedingsof the IEEE International Symposium on Industrial Electronics(ISIE rsquo09) pp 1013ndash1018 IEEE Seoul South Korea July 2009

[18] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) vol 10 pp 25ndash28 Kitakyushu Japan August2014

[19] F Cavallo D Esposito E Rovini et al ldquoPreliminary evalu-ation of SensHand V1 in assessing motor skills performancein Parkinson diseaserdquo in Proceedings of the 2013 IEEE 13thInternational Conference onRehabilitation Robotics (ICORR rsquo13)pp 1ndash6 Seattle Wash USA June 2013

[20] H G Kortier V I Sluiter D Roetenberg and P H VeltinkldquoAssessment of hand kinematics using inertial and magneticsensorsrdquo Journal of NeuroEngineering and Rehabilitation vol 11no 1 article 70 2014

[21] Y Xu Y Wang Y Su and X Zhu ldquoResearch on the calibrationmethod of micro inertial measurement unit for engineeringapplicationrdquo Journal of Sensors vol 2016 Article ID 9108197 11pages 2016

[22] J Keighobadi ldquoFuzzy calibration of a magnetic compass forvehicular applicationsrdquoMechanical Systems and Signal Process-ing vol 25 no 6 pp 1973ndash1987 2011

[23] X Yun and E R Bachmann ldquoDesign implementation andexperimental results of a quaternion-based Kalman filter forhuman body motion trackingrdquo IEEE Transactions on Roboticsvol 22 no 6 pp 1216ndash1227 2006

[24] Xsens Technologies httpwwwxsenscom[25] T Hamel and R Mahony ldquoAttitude estimation on SO(3) based

on direct inertial measurementsrdquo in Proceedings of the IEEEInternational Conference on Robotics and Automation (ICRArsquo06) pp 2170ndash2175 Orlando Fla USA May 2006

[26] S Bonnabel PMartin and P Rouchon ldquoNon-linear symmetry-preserving observers on Lie groupsrdquo IEEE Transactions onAutomatic Control vol 54 no 7 pp 1709ndash1713 2009

[27] InvenSense MPU-9250 Nine-Axis MEMS MotionTrackingDevice 2015 httpwwwinvensensecom-productsmotion-tracking9-axismpu-9250

Scientific Programming 9

0 10 20 30 40minus100

0

100Upper arm

t (s)0 10 20 30 40

minus1

0

1Forearm

t (s)

0 10 20 30 40minus1

0

1Palm

t (s)0 10 20 30 40

minus2

0

2Proximal

t (s)

0 10 20 30 40

0

05Medial

t (s)0 10 20 30 40

minus05minus05

0

05Distal

t (s)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Ang

el (∘

)A

ngel

(∘)

Ang

el (∘

)

Figure 8 The orientations of gestures

2040

6080 minus15 minus1 minus05 0 05 1

minus20

0

20

40

60

y (cm)x (cm)

z (c

m)

StartEnd

Figure 9 The positions of the fingertip of the right forefinger

orientations of gesture are divided into absolute orienta-tions and relative orientations The absolute orientations aredetermined by the QEKFs and the relative orientations areestimated by integrating the kinematics of the arm-handThe

positions are then easily computed The proposed method issimple and fast The algorithm is operated in the embeddedsystem at 50Hz The results of the experiments proved theadvantages of the proposed device

5 Conclusion

This paper presents the design implementation and exper-imental results of a wearable device for gestures captur-ing using inertial and magnetic sensor modules containingorthogonally mounted triads of accelerometers angular ratesensors and magnetometers Different from commercialmotion data gloves which usually use high-cost motion-sensing fibers to acquire hand motion data we adoptedthe low-cost inertial and magnetic sensor to reduce costMeanwhile the performance of the low-cost low-power andlight-weight IMMU is superior to some commercial IMMUFurthermore the novel device is proposed based on thirty-six IMMUs which cover the whole segments of the two armsand hands We designed the online and offline calibrationmethods to improve the accuracy of the units We deducedthe 3D arm and hand motion estimation algorithms thatintegrated the proposed kinematic models of the arm handand fingers and the attitude of gesture and positions offingertips can be determined For real-time performance and

10 Scientific Programming

convenience the interface with virtual model is designedPerformance evaluations verified that the proposed dataglove can accurately capture the motion of gestures Thesystem is developed in the way that all electronic componentscan be integrated and easy to wear This makes it moreconvenient and appealing for the user

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work has been supported by the National Science Foun-dation for Young Scientists of China (Grant no 61503212) andNational Natural Science Foundation of China (Grant nosU1613212 61327809 61210013 and 91420302)

References

[1] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) pp 25ndash28 Kitakyushu Japan August 2014

[2] J M Palacios C Sagues E Montijano and S LlorenteldquoHuman-computer interaction based on hand gestures usingRGB-D sensorsrdquo Sensors vol 13 no 9 pp 11842ndash11860 2013

[3] S S Rautaray and A Agrawal ldquoVision based hand gesturerecognition for human computer interaction a surveyrdquo Artifi-cial Intelligence Review vol 43 no 1 pp 1ndash54 2012

[4] E A Arkenbout J C F de Winter and P Breedveld ldquoRobusthandmotion tracking through data fusion of 5dt data glove andnimble VR kinect camera measurementsrdquo Sensors vol 15 no12 pp 31644ndash31671 2015

[5] D Regazzoni G De Vecchi and C Rizzi ldquoRGB cams vs RGB-D sensors low cost motion capture technologies performancesand limitationsrdquo Journal of Manufacturing Systems vol 33 no4 pp 719ndash728 2014

[6] Y Li X Chen X Zhang K Wang and Z J Wang ldquoA sign-component-based framework for Chinese sign language recog-nition using accelerometer and sEMG datardquo IEEE Transactionson Biomedical Engineering vol 59 no 10 pp 2695ndash2704 2012

[7] L Dipietro A M Sabatini and P Dario ldquoA survey of glove-based systems and their applicationsrdquo IEEE Transactions onSystems Man and Cybernetics Part C Applications and Reviewsvol 38 no 4 pp 461ndash482 2008

[8] B Takacs ldquoHow and why affordable virtual reality shapes thefuture of educationrdquoThe International Journal of Virtual Realityvol 7 no 1 pp 53ndash66 2008

[9] G Saggio ldquoA novel array of flex sensors for a goniometric gloverdquoSensors and Actuators A Physical vol 205 pp 119ndash125 2014

[10] R Gentner and J Classen ldquoDevelopment and evaluation of alow-cost sensor glove for assessment of human finger move-ments in neurophysiological settingsrdquo Journal of NeuroscienceMethods vol 178 no 1 pp 138ndash147 2009

[11] X L Wang and C L Yang ldquoConstructing gyro-free inertialmeasurement unit from dual accelerometers for gesture detec-tionrdquo Sensors amp Transducers vol 171 no 5 pp 134ndash140 2014

[12] W Chou B Fang L Ding X Ma and X Guo ldquoTwo-stepoptimal filter design for the low-cost attitude and headingreference systemsrdquo IET Science Measurement and Technologyvol 7 no 4 pp 240ndash248 2013

[13] D Roetenberg H J Luinge C T M Baten and P H VeltinkldquoCompensation of magnetic disturbances improves inertial andmagnetic sensing of human body segment orientationrdquo IEEETransactions on Neural Systems and Rehabilitation Engineeringvol 13 no 3 pp 395ndash405 2005

[14] J M Lambrecht and R F Kirsch ldquoMiniature low-powerinertial sensors promising technology for implantable motioncapture systemsrdquo IEEE Transactions on Neural Systems andRehabilitation Engineering vol 22 no 6 pp 1138ndash1147 2014

[15] R Xu S L Zhou and W J Li ldquoMEMS accelerometer basednonspecific-user hand gesture recognitionrdquo IEEE Sensors Jour-nal vol 12 no 5 pp 1166ndash1173 2012

[16] E Morganti L Angelini A Adami D Lalanne L Lorenzelliand E Mugellini ldquoA smart watch with embedded sensorsto recognize objects grasps and forearm gesturesrdquo ProcediaEngineering vol 41 pp 1169ndash1175 2012

[17] J-H Kim N DThang and T-S Kim ldquo3-D handmotion track-ing and gesture recognition using a data gloverdquo in Proceedingsof the IEEE International Symposium on Industrial Electronics(ISIE rsquo09) pp 1013ndash1018 IEEE Seoul South Korea July 2009

[18] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) vol 10 pp 25ndash28 Kitakyushu Japan August2014

[19] F Cavallo D Esposito E Rovini et al ldquoPreliminary evalu-ation of SensHand V1 in assessing motor skills performancein Parkinson diseaserdquo in Proceedings of the 2013 IEEE 13thInternational Conference onRehabilitation Robotics (ICORR rsquo13)pp 1ndash6 Seattle Wash USA June 2013

[20] H G Kortier V I Sluiter D Roetenberg and P H VeltinkldquoAssessment of hand kinematics using inertial and magneticsensorsrdquo Journal of NeuroEngineering and Rehabilitation vol 11no 1 article 70 2014

[21] Y Xu Y Wang Y Su and X Zhu ldquoResearch on the calibrationmethod of micro inertial measurement unit for engineeringapplicationrdquo Journal of Sensors vol 2016 Article ID 9108197 11pages 2016

[22] J Keighobadi ldquoFuzzy calibration of a magnetic compass forvehicular applicationsrdquoMechanical Systems and Signal Process-ing vol 25 no 6 pp 1973ndash1987 2011

[23] X Yun and E R Bachmann ldquoDesign implementation andexperimental results of a quaternion-based Kalman filter forhuman body motion trackingrdquo IEEE Transactions on Roboticsvol 22 no 6 pp 1216ndash1227 2006

[24] Xsens Technologies httpwwwxsenscom[25] T Hamel and R Mahony ldquoAttitude estimation on SO(3) based

on direct inertial measurementsrdquo in Proceedings of the IEEEInternational Conference on Robotics and Automation (ICRArsquo06) pp 2170ndash2175 Orlando Fla USA May 2006

[26] S Bonnabel PMartin and P Rouchon ldquoNon-linear symmetry-preserving observers on Lie groupsrdquo IEEE Transactions onAutomatic Control vol 54 no 7 pp 1709ndash1713 2009

[27] InvenSense MPU-9250 Nine-Axis MEMS MotionTrackingDevice 2015 httpwwwinvensensecom-productsmotion-tracking9-axismpu-9250

10 Scientific Programming

convenience the interface with virtual model is designedPerformance evaluations verified that the proposed dataglove can accurately capture the motion of gestures Thesystem is developed in the way that all electronic componentscan be integrated and easy to wear This makes it moreconvenient and appealing for the user

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work has been supported by the National Science Foun-dation for Young Scientists of China (Grant no 61503212) andNational Natural Science Foundation of China (Grant nosU1613212 61327809 61210013 and 91420302)

References

[1] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) pp 25ndash28 Kitakyushu Japan August 2014

[2] J M Palacios C Sagues E Montijano and S LlorenteldquoHuman-computer interaction based on hand gestures usingRGB-D sensorsrdquo Sensors vol 13 no 9 pp 11842ndash11860 2013

[3] S S Rautaray and A Agrawal ldquoVision based hand gesturerecognition for human computer interaction a surveyrdquo Artifi-cial Intelligence Review vol 43 no 1 pp 1ndash54 2012

[4] E A Arkenbout J C F de Winter and P Breedveld ldquoRobusthandmotion tracking through data fusion of 5dt data glove andnimble VR kinect camera measurementsrdquo Sensors vol 15 no12 pp 31644ndash31671 2015

[5] D Regazzoni G De Vecchi and C Rizzi ldquoRGB cams vs RGB-D sensors low cost motion capture technologies performancesand limitationsrdquo Journal of Manufacturing Systems vol 33 no4 pp 719ndash728 2014

[6] Y Li X Chen X Zhang K Wang and Z J Wang ldquoA sign-component-based framework for Chinese sign language recog-nition using accelerometer and sEMG datardquo IEEE Transactionson Biomedical Engineering vol 59 no 10 pp 2695ndash2704 2012

[7] L Dipietro A M Sabatini and P Dario ldquoA survey of glove-based systems and their applicationsrdquo IEEE Transactions onSystems Man and Cybernetics Part C Applications and Reviewsvol 38 no 4 pp 461ndash482 2008

[8] B Takacs ldquoHow and why affordable virtual reality shapes thefuture of educationrdquoThe International Journal of Virtual Realityvol 7 no 1 pp 53ndash66 2008

[9] G Saggio ldquoA novel array of flex sensors for a goniometric gloverdquoSensors and Actuators A Physical vol 205 pp 119ndash125 2014

[10] R Gentner and J Classen ldquoDevelopment and evaluation of alow-cost sensor glove for assessment of human finger move-ments in neurophysiological settingsrdquo Journal of NeuroscienceMethods vol 178 no 1 pp 138ndash147 2009

[11] X L Wang and C L Yang ldquoConstructing gyro-free inertialmeasurement unit from dual accelerometers for gesture detec-tionrdquo Sensors amp Transducers vol 171 no 5 pp 134ndash140 2014

[12] W Chou B Fang L Ding X Ma and X Guo ldquoTwo-stepoptimal filter design for the low-cost attitude and headingreference systemsrdquo IET Science Measurement and Technologyvol 7 no 4 pp 240ndash248 2013

[13] D Roetenberg H J Luinge C T M Baten and P H VeltinkldquoCompensation of magnetic disturbances improves inertial andmagnetic sensing of human body segment orientationrdquo IEEETransactions on Neural Systems and Rehabilitation Engineeringvol 13 no 3 pp 395ndash405 2005

[14] J M Lambrecht and R F Kirsch ldquoMiniature low-powerinertial sensors promising technology for implantable motioncapture systemsrdquo IEEE Transactions on Neural Systems andRehabilitation Engineering vol 22 no 6 pp 1138ndash1147 2014

[15] R Xu S L Zhou and W J Li ldquoMEMS accelerometer basednonspecific-user hand gesture recognitionrdquo IEEE Sensors Jour-nal vol 12 no 5 pp 1166ndash1173 2012

[16] E Morganti L Angelini A Adami D Lalanne L Lorenzelliand E Mugellini ldquoA smart watch with embedded sensorsto recognize objects grasps and forearm gesturesrdquo ProcediaEngineering vol 41 pp 1169ndash1175 2012

[17] J-H Kim N DThang and T-S Kim ldquo3-D handmotion track-ing and gesture recognition using a data gloverdquo in Proceedingsof the IEEE International Symposium on Industrial Electronics(ISIE rsquo09) pp 1013ndash1018 IEEE Seoul South Korea July 2009

[18] B-S Lin I-J Lee P-C Hsiao S-Y Yang and W Chou ldquoDataglove embedded with 6-DOF inertial sensors for hand rehabil-itationrdquo in Proceedings of the 10th International Conference onIntelligent Information Hiding and Multimedia Signal Processing(IIH-MSP rsquo14) vol 10 pp 25ndash28 Kitakyushu Japan August2014

[19] F Cavallo D Esposito E Rovini et al ldquoPreliminary evalu-ation of SensHand V1 in assessing motor skills performancein Parkinson diseaserdquo in Proceedings of the 2013 IEEE 13thInternational Conference onRehabilitation Robotics (ICORR rsquo13)pp 1ndash6 Seattle Wash USA June 2013

[20] H G Kortier V I Sluiter D Roetenberg and P H VeltinkldquoAssessment of hand kinematics using inertial and magneticsensorsrdquo Journal of NeuroEngineering and Rehabilitation vol 11no 1 article 70 2014

[21] Y Xu Y Wang Y Su and X Zhu ldquoResearch on the calibrationmethod of micro inertial measurement unit for engineeringapplicationrdquo Journal of Sensors vol 2016 Article ID 9108197 11pages 2016

[22] J Keighobadi ldquoFuzzy calibration of a magnetic compass forvehicular applicationsrdquoMechanical Systems and Signal Process-ing vol 25 no 6 pp 1973ndash1987 2011

[23] X Yun and E R Bachmann ldquoDesign implementation andexperimental results of a quaternion-based Kalman filter forhuman body motion trackingrdquo IEEE Transactions on Roboticsvol 22 no 6 pp 1216ndash1227 2006

[24] Xsens Technologies httpwwwxsenscom[25] T Hamel and R Mahony ldquoAttitude estimation on SO(3) based

on direct inertial measurementsrdquo in Proceedings of the IEEEInternational Conference on Robotics and Automation (ICRArsquo06) pp 2170ndash2175 Orlando Fla USA May 2006

[26] S Bonnabel PMartin and P Rouchon ldquoNon-linear symmetry-preserving observers on Lie groupsrdquo IEEE Transactions onAutomatic Control vol 54 no 7 pp 1709ndash1713 2009

[27] InvenSense MPU-9250 Nine-Axis MEMS MotionTrackingDevice 2015 httpwwwinvensensecom-productsmotion-tracking9-axismpu-9250

Scientific Programming 11

[28] M Chen S Gonzalez A Vasilakos H Cao and V C MLeung ldquoBody area networks a surveyrdquo Mobile Networks andApplications vol 16 no 2 pp 171ndash193 2011

[29] MIT Media LabMIThril Hardware Platform 2015 httpwwwmediamiteduwearablesmithrilhardwareindexhtml

[30] S Salehi G Bleser N Schmitz and D Stricker ldquoA low-costand light-weight motion tracking suitrdquo in Proceedings of theIEEE 10th International Conference on Ubiquitous Intelligenceand Computing and IEEE 10th International Conference onAutonomic and Trusted Computing (UICATC rsquo13) December2013

[31] J FangH Sun J Cao X Zhang andY Tao ldquoA novel calibrationmethod of magnetic compass based on ellipsoid fittingrdquo IEEETransactions on Instrumentation and Measurement vol 60 no6 pp 2053ndash2061 2011

Submit your manuscripts athttpswwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Distributed Sensor Networks

International Journal of

Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014

International Journal of

ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia

International Journal of

Biomedical Imaging

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ArtificialNeural Systems

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Computational Intelligence and Neuroscience

Industrial EngineeringJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014