Robust Walking Direction Estimation via Principal Component Pursuit in Pedestrian Dead Reckoning

In pedestrian dead reckoning (PDR) systems, accurate estimation of the walking direction is of paramount importance for achieving high localization accuracy. Despite numerous research endeavors in this area, a remaining challenge lies in cases where the direction indicated by the sensors does not align with the actual pedestrian's walking direction. In this paper, we address this misalignment problem by focusing on the PCA-based method. While this method exploits the advantages of PCA, it is also subject to its limitations, notably its susceptibility to outliers due to abnormal actions beyond normal walking, which are commonly encountered in PDR scenarios. To ensure robustness against outliers, we propose a novel PCA-based method for estimating the walking direction. Through experiments conducted on various actions, we evaluate the performance of the proposed method. The results demonstrate that our proposed method, which applies stable principal component pursuit for robust PCA, achieves an improved performance with an RMSE of 5.15° compared to classical PCA. Furthermore, the proposed method exhibits stable estimation and robustness across all presented actions and sequences.