Robust pedestrian dead reckoning using anchor point recalibration

All indoor positioning approaches face the challenge to deal with erroneous input data from sensors. Especially independent systems like dead reckoning rely on highly accurate input data from accelerometers and gyroscopes for an accurate prediction of the user's position. But with the widespread of affordable mobile devices equipped with low-cost sensors, the obtained input data is noisy and of poor quality. Since errors accumulate within a pedestrian dead reckoning (PDR) system, there is an inevitable need for recalibration on a regular basis. We propose a PDR system based on state-of-the-art particle filters, which is recalibrated using both anchor points and a pedestrian movement model. Our evaluation compares standard particle filters with a backtracking particle filter including information from indoor maps and our enhancements. We show that a combination of anchor point recalibration, error calculation of sensor bias, and a fine-tuned movement model can decrease the RMSE by 1.07 m.