IOP PUBLISHING JOURNAL OF GEOPHYSICS AND ENGINEERING

J. Geophys. Eng. 7 (2010) 404–409 doi:10.1088/1742-2132/7/4/007

Using an optical mouse sensor to trackgeophysical field measurements

Reinhard Klose and Jorg Schmalzl

Institut fur Geophysik, WWU Munster, Corrensstrasse 24, D-48149 Munster, Germany

E-mail: joergs@uni-muenster.de

Received 16 March 2010Accepted for publication 3 August 2010Published 1 October 2010Online at stacks.iop.org/JGE/7/404

Abstract

To position arbitrary outdoor measurements, the most common methods are the use of globalpositioning system (GPS) technology or distance tracking from reference locations. Thesystem to be introduced here consists of a combination of an optical distance tracking systemand a civil, autonomous GPS receiver. Through the application of a recursive filter, bothcomponents support and correct each other mutually. The system is intended as anautonomous, real-time and low-cost alternative to expensive differential GPS solutions ingeophysical field surveying. The combination approach features a very good relative anddecent absolute positioning precision. First measurements using the two-coilgeo-electromagnetics instrument Geonics EM31 in combination with the developedpositioning system show promising results.

Keywords: environmental science, Earth science, instrumentation, measurement

Introduction

For the economy of ground-based, small-scale geophysicalfield surveying, rapid, high-quality data acquisition isdesirable. Therefore, not only have adequate instruments tobe utilized, but also a precise and fast data positioning methodis essential.

Position information can be obtained by several methods.The most common are distance tracking from referencelocations or the use of GPS (global positioning system)technology.

Tracking wheel odometers are precise, yet positioning isrelative, as the quantity being measured is a distance, not asabsolute point in space. Furthermore, measurement errorsaccumulate with covered distance and, due to slippage or lossof contact, they cannot be used on any ground.

GPS position tracking using autonomous, civil GPSreceivers is absolute, but it is, for the application in geophysicalsurveying, generally not accurate enough. To improve GPSpositioning, expensive (both in purchase and work effort)differential GPS (DGPS) solutions have to be applied.

The positioning system to be introduced here consistsof a combination of a common GPS receiver and a two-dimensional odometer based on an optical mouse sensor. Bothsystems mutually support each other, as GPS position integrity

is improved by the additional displacement information andaccumulating errors in distance tracking are compensated bythe incorporation of absolute GPS positions.

Optical flow tracking

Besides their obvious deployment in optical computer mice,optical mouse sensors are extensively used as two-dimensionalodometers in the field of mobile robotics (Palacin et al 2005,Baek et al 2005, Dille et al 2009). To use this technique also ingeophysical surveying applications, a prototype of an outdooroptical tracking system is introduced here.

An optical mouse sensor in principle consists of a low-resolution camera with a high frame rate and an imageprocessing unit. Here, the Agilent ADNS3080 (AgilentTechnologies GmbH 2005) high-performance optical mousesensor (figure 1) is used. By comparing consecutive framesof a sequence, the image processor can determine the two-dimensional projection of, in general, three-dimensionalrelative displacements between the camera and the observedsurface. To do this, sensor algorithms compute the opticalflow on the camera image plane. Output information is adisplacement (dx, dy) in units of pixels.

Optical flow is a two-dimensional vector field in the imageplane of the camera. A vector proportional to the displacement

1742-2132/10/040404+06$30.00 © 2010 Nanjing Geophysical Research Institute Printed in the UK 404

Using an optical mouse sensor to track geophysical field measurements

Figure 1. The Agilent ADNS3080 optical mouse sensor (left)features a frame rate of 6400 fps. A sensor frame (right) has aresolution of 30 × 30 pixels (Agilent Technologies GmbH 2005).The depicted image is made with the sensor being mounted on aCCTV lens with an elevation of 60 cm above a computer keyboard.

(a) (b) (c)

Figure 2. Schematic of the optical flow (c) calculated from twoconsecutive frames (a) and (b) of a sequence. Indicated are atranslation (top) and a rotation (bottom) of subimages.

between consecutive frames of a pixel’s intensity value iscomputed for every pixel on the image plane (figure 2). Usingoptical flow algorithms (Horn and Schunck 1980, Lucas andKanade 1980, Beauchemin and Barron 1995), rotations andtranslations of images or subimages can thus be determined.For the application in optical mice, only a global translationof the whole image is of interest.

Due to the assumed operation on even surfaces close tothe sensor lens, the standard optical mouse configuration is notpractical in outdoor applications. Therefore, the optical sensoris mounted on a 16 mm CCTV (close circuit television) cameralens to to able to increase its elevation and thus its footprintsize. To credibly detect displacements, the observed surfacehas to be sufficiently textured. Otherwise, algorithms cannotkeep track of recognizable features in consecutive frames. As ameasure of intensity variance within an image, the ADNS3080reports a surface quality value. Experience shows that thesensor, in combination with the camera lens, outdoors and indaylight detects sufficiently high surface quality values on anyground to reliably determine displacements.

To transform the sensor’s displacement in units ofpixels into a real-world displacement in units of metres,information on the distance from the lens to the observedsurface has to be available because the footprint size of acamera normal to a surface and thus the ‘real size’ of apixel on its image plane depends linearly on that distance(figures 3(a) and (b)). Therefore, the ultrasonic rangerDevantech SRF021 is installed next to the optical sensor. It

1 http://www.robot-electronics.co.uk/htm/srf02techI2C.htm

(a)

(b)(c)

Figure 3. Dependence of the optical mouse sensor’s displacementmeasurements on its alignment relative to the surface. Moving thesensor in alignments (a), (b) and (c) at a given distance parallel tothe surface would result in three different displacementmeasurements. Additional distance to the surface and orientationmeasurements thus are necessary to correct the measured values.

Figure 4. Dependence of the conversion factor s on height h, whichtransfers the optical sensor’s displacement from units of pixels tounits of metres.

determines the distance to a reflector by measuring the sonicpulse echo runtime. A height-dependent conversion factor s,which maps pixel displacements to real-world distances, isthen assessed experimentally.

Sensors are installed strictly horizontal on a mobile deviceand for different sensor heights h, a known straight distance iscovered. The known distance is divided by the optical sensor’spixel displacement and plotted against the height. The desiredconversion function is the interpolation to the measurements(figure 4).

Displacements can now be expressed in units of metres,yet only in a local, sensor-defined coordinate system.Furthermore, the sensor is unable to detect any rotations. Toovercome this problem, the three-dimensional magnetic fieldand acceleration sensor Honeywell HMC6343 (HoneywellInternational Inc., 2008) is installed. This sensor is able todetect a tilt compensated compass bearing and pitch and rollangles. With this information, on the one hand, errors resultingfrom image distortions due to sensor dipping (figure 3(c))can be computed and compensated. On the other hand, thesensor’s displacement (dx, dy) in its local coordinate systemcan be transferred to a geographic displacement (de, dn) in aCartesian geographic map projection (figure 5):

dn = dy · cos(bearing) − dx · sin(bearing),

405

R Klose and J Schmalzl

Figure 5. Mapping the optical sensors’s local coordinate system toa geographic, Cartesian projection.

(a) (b) (c)

Figure 6. The installation of the sensors. (a) Bottom: an ultrasonicranger and a camera lens. (b) Top: display for the state of health,displacement, distance to surface and orientation information.(c) Inside view with sensors and a microcontroller. Themicrocontroller transfers sensor data to an USB port.

de = dy · sin(bearing) + dx · cos(bearing).

With the applied sensors, it is now possible to determinea rotation-independent displacement in real-world units on ageographic map projection’s coordinate system (the magneticnorth direction in a first approximation is assumed to be thesame as the geographic north direction). The installation ofthe sensors is depicted in figure 6.

The precision of displacement measurements dependson the installation of the sensors. With firmly horizontaladjustment in constant height and orientation on a mobiledevice and on a straight profile (compass sensor thus notneeded), errors are well beyond 1%. When manuallyand freely carried around, variations in the sensor’s heightand orientation are inevitable. Errors detected by firsttest measurements in that configuration are in the order ofabout 6%.

GPS tracking

Position determination with civil and autonomous GPSreceivers inherits biases of several metres. Figure 7 shows

Figure 7. Test measurements with a stationary SiRFstarIII GPSreceiver. The timespan covered is 16 h with a record interval of 15 s.The mean value has been subtracted from all records. The standarddeviation is 2.02 m and the maximum deviation is 10.04 m.

the details of a test measurement carried out with a modernSiRFStarIII GPS receiver. For the whole timespan of thatmeasurement, the receiver remained stationary.

As a stand-alone positioning system for use in geophysicalfield surveying, these receivers, in general, are too inaccurate.The major drawback here is the error-prone relativepositioning of consecutive GPS measurements (see alsofigure 9). An interpolation calculation on associatedgeophysical field measurements will thus weight adjacent datapoints inaccurately. This leads to blurred out structures in theobtained maps (figure 10). To improve positioning precision,additional spatial information provided by secondary sourcesis inevitable. This information may originate from other GPSreceivers (DGPS), or from any system gathering relative orabsolute position data.

DGPS methods use two or more spatially proximate GPSreceivers. Significant errors in position determination, forexample the ionospheric signal runtime delays or satelliteephemeris and clock errors, affect these receivers in a similarmanner. Stationary base receivers on known positions arethus able to calculate correction data with which positionsacquired by rover receivers are adjusted (El-Rabbany 2002).Correction data are incorporated either in real time or in post-processing. Real-time differential solutions rely on expensivecommunication hardware and constant signal availability.Post-processing is time consuming and data visualizations arenot possible until single records are spatialized.

As an alternative solution, here, optical distance trackingis to be introduced as a secondary spatial information sourceto improve GPS positioning.

406

Using an optical mouse sensor to track geophysical field measurements

Combination of GPS and optical flow sensors

GPS measurements of autonomous, civil receivers areabsolute, yet insufficiently precise. Measurement errors of theoptical distance tracking system are relatively low for smalldisplacements, but grow considerably with covered distance.When combined, on the one hand, GPS position integrity canbe improved by the additional displacement information. Onthe other hand, relative errors in distance tracking can becompensated by the incorporation of absolute GPS positioninformation.

To combine measurements from the GPS and an opticalsensor, both have to share a single-coordinate system. Withthe aid of the electronic compass, optical sensor displacementis available in a geographic, Cartesian coordinate system.Most common GPS receivers report position information inthe NMEA-0183 format. Given that output, a location isspecified by geographic latitude and longitude in degrees. Tomap the geographic latitude and longitude to a geographic andCartesian coordinate system, such as for example the UTM(Universal Transverse Mercator) system, there are commonlydefined projection algorithms available (Kienast et al 1994).Displacements and GPS positions are combined here using arecursive filtering process.

The filtering approach to be introduced here sharesthe predictor–corrector and data-weighting ideas of thestatistically optimal Kalman filtering (Maybeck 1979, Welchand Bishop 2006, Zarchan 2005).

Based on a starting position �P0 = (e0, n0), the opticaldisplacement �V = (de, dn) is used to predict a new positionand a weighted GPS measurement �G = (eg, ng) correctsthat position. The weight is determined by the relativemeasurement errors of displacement �D and GPS �G.

The GPS error is assumed to be constant and twicethe standard deviation (95% of measurements lie within thatradius) of the test measurement introduced in figure 7:

�G = 4.04 m.

The displacement precision is assumed to be 6%. Its absolutevalue depends on the covered distance D = | �V | between twoGPS measurements:

D =√

de2 + dn2 �⇒ �D = 0.06 ·√

de2 + dn2.

Given that, the weight σ can be determined as

σ = �D

�D + �G,

and the recursive filtering equation becomes

�Pk = ( �Pk−1 + �V )︸ ︷︷ ︸

predict

+ σk[ �Gk − ( �Pk−1 + �V )]︸ ︷︷ ︸

correct

.

Application example

For first performance tests, the newly developed positioningsystem is used in combination with the Geonics EM31. TheEM31 is a two-coil geo-electromagnetics instrument. Theoptical tracking system is mounted on the instrument, together

Figure 8. The optical positioning system and the palmtop computermounted on the Geonics EM31.

with a mobile palmtop computer with an internal GPS receiver(figure 8).

The EM31 is mostly used for contact-free and fastdetection of lateral ground conductivity variations near to thesurface. The instrument can be operated by a single person. Itis carried by the side at hip height by a shoulder strap (GeonicsLtd 1995).

The palmtop is connected to the EM31 and the opticaltracking system via USB, reading both the device’s sensordata and its internal GPS data. During field measurements,the palmtop continuously computes the current position inreal time based on the above filtering equations and logs ittogether with current EM31 measurements to its memory.

As the optical system is mounted on the EM31 andcarried at the operator’s side, fluctuations in sensor alignmentare inevitable. Results of a first test measurement of thatcombination approach are depicted in figure 9.

Nevertheless, significant improvements are made,especially in relative positioning and even perpendicular tothe profiles. With support of the optical tracking system, thus,considerably large fluctuations in the raw GPS measurementscan be adjusted. The effects can be seen in interpolationmaps of EM31 measurements originating at another testmeasurement on the military airbase in Fassberg, Germany.Figure 10 shows the same interpolation map section calculatedfor the same data set, where on the left data are positioned usingthe GPS alone and on the right supported by displacementinformation through the above filtering equation. Resolutionof structures can be improved significantly and for manyapplications is thus sufficiently high.

The integrity of the newly developed positioning methodis verified by the successful return to the starting position(figure 9). Here only a small deviance is recognizable.Yet, the absolute positioning precision depends to a largeextent on the precision of the starting value in the filteringprocess. In this test measurement, the starting value has been

407

R Klose and J Schmalzl

Figure 9. Test measurement with the combination approach ofoptical distance tracking and GPS positioning. North–south facing,parallel profiles with a spacing of 2 m and a length of 50 m are laidout. While covering these, sensors are carried at hip height mountedon the EM31. Raw GPS positions (green) and filtered positions(red) are logged with a frequency of 1 Hz.

determined by a short-term average of 50 GPS measurements.Nevertheless, absolute precision is assumed to be only slightlybetter than that of a civil GPS receiver. For more convenientand more quantitative information on system precision andits dependence on, for example, surface, terrain, illumination,sensor fluctuations or GPS reception, further research will bedone.

Conclusion

As a positioning system for geophysical field surveying, acombination of an optical distance tracking system and a civil,autonomous GPS receiver has been developed. Through theapplication of a recursive filtering approach, both componentssupport and correct each other mutually.

It is shown that the combination approach features a verygood relative and a decent absolute positioning precision, evenunder considerable fluctuations in sensor alignment. Thereached measurement resolution is, for many applications, thussufficiently high.

The system is intended as an autonomous and low-cost(the costs of the applied sensors, a microcontroller and aGPS receiver are about €300) alternative to expensive DGPSsolutions in the field of mobile and small-scale geophysicalsurveying, especially in combination with the EM31. Withoutgoing into detail, there is a wide range of applicationsimaginable, in which DGPS corrections are not availableand the provided stand-alone GPS positioning is insufficientlyprecise.

Furthermore, if alignment fluctuations are reduced andthe compass usage (and its biases) can be avoided, an

Figure 10. Interpolation maps of EM31 measurements (calculatedwith ESRI ArcMap’s spline interpolation). Both maps show thesame map section and the same data, yet positioned by differentmethods. With data positioned by the GPS in combination withdisplacement information (right) a much higher resolution ofstructures is achieved than with data positioned by the GPSalone (left).

application as ‘optical tracking wheel’ is self-suggesting.Possibly, autonomous tracking, with a precision in the orderof conventional tracking wheel precisions, is achievable andthe system can be used for measurement-triggering. Theadvantages here are evident, as optical tracking is contact-freeand applicable even on grounds like, for example, fine sand orice. Using focused artificial illumination, optical tracking caneven be utilized in darkness.

Considering the wide range of possible applications,further research will be done. Particularly, error analysis andtargeted improvement of single components are the next steps.To be able to make more convenient statements on precision,further performance tests are planned.

References

Agilent Technologies GmbH 2005 ADNS-3080 High-performanceOptical Mouse Sensor Data-Sheet

Baek S, Park H and Lee S 2005 Mobile robot localization based onconsecutive range sensor scanning and optical flowmeasurements Proc. 12th Conf. on Advanced Robotics(Department of Mechanical and System Design Engineering,Hongik University, Seoul)

Beauchemin S S and Barron J L 1995 The computation of opticalflow ACM Computing Surveys (CSUR) (London: Departmentof Computer Science, University of Western Ontario)

408

Using an optical mouse sensor to track geophysical field measurements

Dille M, Grocholsky B and Singh S 2009 Outdoor downward-facing optical flow odometry with commodity sensorsProc. Field and Service Robotics (FSR ’09) (Robotics Institute,Carnegie Mellon University, Pittsburgh)

El-Rabbany A 2002 Introduction to GPS: The Global PositioningSystem (Boston, MA: Artech House)

Geonics Ltd 1995 Geonics EM31-MK2 Operating ManualHoneywell International Inc. 2008 HMC6343 3-Axis Compass with

Algorithms Data-SheetHorn B K P and Schunck B G 1980 Determining Optical Flow

(Artificial Intelligence Laboratory) (Cambridge, MA:Massachusetts Institute of Technology)

Kienast G, Hofmann-Wellenhof B and Lichtenegger H 1994 GPS inder Praxis, Abteilung fur Landesvermessung undLandinformation, Technische Universitat Graz (Berlin:Springer)

Lucas B D and Kanade T 1980 An iterative image registrationtechnique with an application to stereo vision Proc.7th Int. Joint Conf. on Artificial Intelligence (ComputerScience Department, Carnegie-Mellon University,Pittsburgh)

Palacin J, Valganon I and Pernia R 2005 The optical mouse forindoor mobile robot odometry measurement SensorsActuators 126 141–7

Maybeck P S 1979 Stochastic Models, Estimation and Control vol 1(New York: Academic) chapter 1

Welch G and Bishop G 2006 An Introduction to the Kalman Filter(Chapel Hill, NC: Department of Computer Science,University of North Carolina)

Zarchan P 2005 Fundamentals of Kalman Filtering—A PracticalApproach (Lexington: Lincoln Laboratory, MassachusettsInstitute of Technology)

409