Indoor Localization with Particle Filter in Multiple Motion Patterns

In this paper, a novel mobile tracking method based on pedestrian dead reckoning (PDR) and wireless local area network (WLAN) RSS fingerprint is proposed, which estimates the real-time location of pedestrian continuously using the improved particle filter. The existing PDR systems mostly focus on the condition that sensor axes are relatively fixed to user. However, the sensor axes may be changing during walking period in several motion patterns, for example that the smartphone is swinging with hand or kept in bag. Therefore, we propose a novel PDR algorithm for different handheld patterns, which detects the steps based on multimode finite-state machine (MFSM) with adaptive updating thresholds and estimates the heading direction with principal component analysis (PCA) and ambiguity resolution. On the other hand, for a long continuous walking process, localization error will accumulate and lead to the particle filter losing tracking of the target device. To deal with this problem, we design an improved particle filter with advanced resampling strategy for recovery when localization fails. We conduct experiments for four motion patterns in two realistic representative indoor environments: office building and shopping mall. Experiment results reveal the proposed localization system could achieve an average localization accuracy within 2m even in the toughest motion pattern.