Outlier Filtering Algorithm for Indoor Pedestrian Walking Direction Estimation

Abstract

This paper introduces an outlier filtering algorithm to improve the indoor pedestrian walking direction estimation accuracy performance. Our previous proposed RMPCA approach combines rotation matrix (RM) and Principal Component Analysis (PCA) to extract pedestrian walking direction using a smartphone in the trouser pocket. Performance of the RMPCA approach may deteriorate if an irregular leg locomotion occurs or device slides in the pocket. If this situation occurs, it may be detected by the proposed outlier filtering algorithm. Then, walking direction of the current step may be obtained by averaging the walking direction estimations of the adjacent normal walking steps. Experiments show that the proposed outlier filtering algorithm may avoid large estimation errors and improve accuracy performance of RMPCA approach.

Foundation Item: This work was supported by the National Natural Science Foundation of China (Granted Nos. 61301132, 61300188, and 61301131), Natural Science Foundation of Liaoning Province of China No. 201602073, and the Fundamental Research Funds for the Central Universities Nos. 3132017129 and 3132016347.