improving pedestrian tracking accuracy with a low-cost imu ... · inertial navigation position...
TRANSCRIPT
Improving Pedestrian Tracking Accuracy With a
Low-cost IMU Through the Use of Tightly-
coupled L1 GPS
Stanley Radzevicius, W. Todd Faulkner, David W.A. Taylor
ENSCO, Inc.
WPI PPL Workshop 2010
August 2, 2010
© 2010 ENSCO, Inc. All rights reserved.
Unobstructed GPS
Introduction
Tracking first responders requires navigation in GPS-
impaired environments.
IMU is self contained and provides good short-term
navigation, but errors increase with time.
GPS requires sky view, suffers from multipath, and can be
jammed but provides meter-level accuracy when
available.
Integrating GPS extends tracking capability of standalone
boot-mounted IMU/compass mission time in GPS-
impaired environments.
2
Coverage Gaps
GPS-Denied
© 2010 ENSCO, Inc. All rights reserved.
Boot-mounted IMU and Compass
Unaided inertial navigation position errors grow as time cubed with gyro biases.
Through an extended Kalman filter (EKF),
1. Zero-velocity updates (ZUPTs) constrain position errors.
2. Heading updates using filtered compass data constrain heading errors; updates
are typically available even in magnetically complex indoor environments.
Push-OffSwing Heel-Strike
Stance
ZUPT
Positioning accuracy of a few meters for up to 15 minutes.
3
© 2010 ENSCO, Inc. All rights reserved.
Why integrate GPS with a boot-mounted IMU?
Complementary errors:
GPS position errors are larger but do not
grow with time.
Inertial navigation position errors are more
accurate over short periods of time but
drift over long periods.
Initialization:
GPS provides an initial absolute position
estimate and can be used to estimate
inertial heading, magnetometer and IMU
errors, especially accelerometer biases,
as the operator walks.
GPS Measurement Filtering:
Inertial position estimates help reject poor
GPS data.
4
© 2010 ENSCO, Inc. All rights reserved.
Loosely Coupled GPS/IMU: Two Separate EKFs
GPS
Sensor
GPS
Kalman
Filter
Pseudoranges
& Doppler
Error States:
3 Position
3 Velocity
Clock Bias
Clock Drift
IMU
Sensor
Inertial
Navigation &
Kalman
Filter
Incremental Angle
& Velocity
Error States:
3 Position
3 Velocity
3 Attitude
3 Gyro Bias
3 Acc Bias
Position
Velocity
Position
Velocity
Attitude
5
© 2010 ENSCO, Inc. All rights reserved.
Tightly Coupled GPS/IMU: Single EKF
IMU
Sensor
GPS/IMU
Inertial
Navigation &
Kalman Filter
Incremental Angle
& Velocity
Error States
3 Position
3 Velocity
3 Attitude
3 Gyro Bias
3 Acc Bias
Clock Bias
Clock Drift
Position
Velocity
Attitude
GPS
Sensor
Pseudoranges
& Doppler
6
© 2010 ENSCO, Inc. All rights reserved.
Why use a tightly coupled approach?
Chi-square goodness of fit (Χ2) test checks for outlier measurements.
Loosely coupled approach: GPS position estimates input to the
inertial EKF may be based on correlated pseudorange errors; Χ2 test
accepts/rejects a position measurement, even if based on a single
bad pseudorange.
Tightly-coupled approach: Χ2 test accepts/rejects each pseudorange
measurement independently (it may reject a single bad pseudorange
measurement while accepting adjacent measurements). This
approach also allows integration of sparse pseudorange
measurements that are discarded in loosely coupled approach.
7
© 2010 ENSCO, Inc. All rights reserved.
System Components
MEMSense nIMU ~ $1200
IMU & Compass
Size: 0.8 cubic inches
Weight: < 1 ounce
Power: < 1 W
Gyro Bias Instability: 70°/hr
u-blox 5 GPS receiver
and u-blox GPS antenna
Experimental Hardware
8
© 2010 ENSCO, Inc. All rights reserved.
GPS/Boot-mounted IMU Test Results
Operator walking for over an hour and
covering multiple miles.
GPS coverage environments:
1. Unobstructed with continuous GPS
coverage
2. Obstructed environment with
intermittent GPS coverage
3. GPS-denied indoor environment –
primarily inertial navigation solution
Position results overlaid on Google Earth
imagery for ground truth.
9
© 2010 ENSCO, Inc. All rights reserved.
Indoor/Outdoor Walk (65 minutes, 2.75 miles)
Intermittent dropouts
Multipath reflectors Multipath & partially
GPS-denied
0.5
mile
s
© 2010 ENSCO, Inc. All rights reserved.
Indoor/Outdoor Walk: GPS Only
11
Large GPS
ErrorsPoor Indoor Navigation
Loss of GPS
© 2010 ENSCO, Inc. All rights reserved.
Indoor/Outdoor Walk: Tightly Coupled GPS/IMU
GPS Multipath Errors
Greatly Reduced
Good Indoor Navigation
12
© 2010 ENSCO, Inc. All rights reserved.
Indoor/Outdoor Tightly Coupled GPS/INS
Navigation: Short Segment
Enter Building
Exit Building
GPS Only GPS/IMU
13
© 2010 ENSCO, Inc. All rights reserved.
70 Minute Walk Under Tree Canopy
4.5 mile walk around lake
Tree Canopy for > 75% of walk
Only lost GPS for 5 minutes
14
© 2010 ENSCO, Inc. All rights reserved.
Canopy Walk: GPS Only Solution
7-10 satellites available
except for 5 minute
period where no
satellites were visible.
15
© 2010 ENSCO, Inc. All rights reserved.
Real time Processing: IMU Bridges 5-minute GPS Gap
Using only the boot-mounted IMU,
position estimation remains
consistent with overhead
imagery of the trail after loss of
GPS for 5 minutes.
16
200 m
Red = GPS
Green = Tightly coupled GPS/IMU
Black line = Trail location in imagery
© 2010 ENSCO, Inc. All rights reserved.
IMU bridges 30-minute GPS Gap
Red = GPS
Blue = GPS/IMU solution
Real-time position error based on imagery after 30 minutes without GPS
(covering 1.75 miles) is ~15 meters
17
Start of GPS outage
End of GPS outage
© 2010 ENSCO, Inc. All rights reserved.
Conclusions
Tracking first responder for extended periods of time
without GPS is feasible using a boot-mounted IMU
and compass.
However, even sparse GPS data in a GPS/IMU
system provides:
Initialization of position, heading and IMU sensor
errors for inertial navigation
Correction of long-term inertial navigation errors
18
