[Summary Week4] Pedestrian tracking with Shoe Mounted Inertial sensors – Eric Foxlin

In this paper, the author describes NavShoe, a novel shoe-mounted inertial sensor system that is capable of tracking 6-DOF in an un-instrumented environment. The proposed system builds on the prevalent MEMS gyroscopes that report 3-DOF (orientation in space) by also providing highly accurate position estimates.

Un-instrumented inertial tracking is made feasible by the identification of alternating stationary and moving stride phases of human locomotion. The inertial sensors, being shoe-mounted, can detect this pattern. The stationary phases are used to apply zero-velocity updates (ZUPTs), to ensure that the inertial integrator’s velocity is set to zero intermittently. Another novel aspect is the feeding of ZUPTs to an Extended Kalhman Filter(EKF) to correct velocity errors, leading to error accumulation that is linear in strides as opposed to cubic. The ZUPT feedback to the EKF is able to correct for all drifts, barring yaw drift. The author minimizes this heading drift using corrective measurements obtained from a magnetic compass.

To realize the proposed system, the author employs triaxialrate-gyroscopes, accelerometers and magnetometers. To drive real-world consumption, the system is also provided with a RF transceiver that receives sensor information which is then processed by a host of stochastic prediction/correlation algorithms which yield the corrected position and orientation.

The author describes how compensating for magnetic interference was a major problem for him to solve, and discusses a scheme that allows the system to be calibrated initially and thus be subsequently free of magnetic-field distortions created by metal in shoe-soles. He also elaborates on the technique that he uses for modeling the magnetic compass output that accounts for noise as well as spatial magnetic field variations due to declination and deviation.

Finally, the author present his results of the tracking system in an indoor and an outdoor setting. The results are quite impressive, and there is only a 0.3% heading error in both settings. The author also describes a possible hybrid system that could incorporate GPS data as a position-correction term, thus making the system viable even for long treks. As extensions to his work, the author describes the possibility of using additional spatial information for improving accuracy (much akin to the GeoSpots of Argon), and also the potential to use the system for driving a HMD.

[week 4 summaries]

Pedestrian Tracking with Shoe-Mounted Inertial Sensors

There are a bunch of scenarios when a navigation for tracking the location of a person is useful, some are for life saving (locate to rescue firefighters); some are for assistant (personal navigation); and some are for entertaining (AR applications).

The author put forward a pedestrian tracking system NavShoe that leverages inertial sensors for position tracking. It is a compact wireless sensor tuck into the shoelaces and works both for indoor and outdoor applications. The inertial sensor, unlike other tracking approaches which need installed markers or prior instrumentation, can works within a reasonable accuracy in arbitrary unprepared environment. Furthermore, the new system provides an extremely accurate orientation tracker on foot.

A big problem which affects the accuracy of inertial sensor is the accumulated drift error. The NavShoe tries to cope with integrating error by introducing zero-velocity updates (ZUPTs) into the Extended Kalman Filter (EKF). As people walks, his gesture can be classified into two phases: stationary stance phase and moving stride phase. The NavShoe detects the stance phase and applies ZUPT as measurement to break the error accumulation from cubic-in-time to linear in the number of steps.

Another problem comes from the deviation of compass. The distortion of the earth’s magnetic field caused by the metal in the shoes is overcome by a delicate calibration before entering navigation mode. This calibration only needs to be done when we remove the sensor to another shoe.

A bunch of test results are given covering indoor/outdoor experiments and those integrated with GPS’s. The indoor experiment takes 3 shots with predefined route; free-trail (but mostly following the first route) and route involving altitude changes. The results shows good performance with 0.3 percent of the distance traveled in plane and 0.6 in altitude. The experiment of outdoors also obtains similar good performance. By integrated with GPS, NavShoe can deal with ever longer distance but still maintain the accuracy by online calibration of magnetic declination.

In the future, we can also using the foot sensor to increase the head tracking sensor’s accuracy.

Overall, I think this system can be a complement for GPS’s inaccuracy in small scale and dependency on pre-mapped environment. Scenarios that need fine position data or have limited access to the satellite system give this application places in the market.

Week 4 Summary

Navshoe is new kind of tracking technology that is used to track motion of a user in the real world.  The basic design is small device with various sensors is placed in the shoe of the user.  The device is GPS capable but by now means requires GPS to function.  It combines a few different sensors together being a gyro, accelerometer and an inertia senors.  Unlike a pedometer the device actually tracks the displacement of the user in three dimensions.  It is meant for indoors and outdoors.  The device uses Kalman filtering to reduce the noise coming from the sensors.

To improve accuracy, uses data from all available sensors to reduce the noise level in it’s calculation.  Even GPS isn’t perfect but the system can recognize sudden GPS jumps by comparing the GPS data to that of the other part of the device.  GPS is used to reduce the error in the system by occasionally providing landmarks.  The NavShoe does occasionally give bad readings.  To combat long term error the devices uses pauses in movement of the users foot to fix the location elimanting the long term build up error instead allowing small amount that build liner of over great distances.

Overall I think it;s an interesting idea.  Even more so when combined combined with an existing system such as a HMD or some kind of vision system.