Robust Adaptive Extended Kalman Filtering for Smart Phone-based Pedestrian Dead Reckoning Systems

Abstract

The expansion of Location-based services and applications leads to extensive interests on smart phone-based indoor and outdoor localization. Rich sensors embedded in smart phone support varies of localization techniques, provide infrastructural elements for indoor and outdoor seamless localization solutions. The pedestrian dead reckoning (PDR) system based on smart phone-embedded MEMS sensors plays an important role in a seamless localization system, since it can link up different absolute positioning systems (such as BeiDou Navigation Satellite System (BDS), WiFi localization systems, etc.) flexibly. However, as a relative localization system, it is limited to location error accumulation, and therefore it cannot run for long. The problem can also affect the performance of a seamless localization system. As a result, in order to improve the tracking performance of the PDR system in complex environments indoors and outdoors, a method based on Robust Adaptive Extended Kalman Filtering (RAEKF) is proposed. The method includes heading and speed estimation, for heading estimation, outputs from gyroscope, accelerometer, and magnetometer sensors are used, and for speed estimation, only outputs from accelerometer are used. RAEKF is employed both in heading and location estimation. Although speed and location estimation refer to different state and measuring models, the proposed filtering can be applied flexibly. The M-estimator is used to handle measurement outliers. To weaken the impacts of dynamic disturbance errors for heading and location estimation, an adaptive factor is introduced to adjust their models respectively. Extensive experiments on static and dynamic localization were conducted in indoor complex environments. And the experimental results demonstrate the proposed method provides more accurate and robust performances, compared with methods based on conventional EKF.