Mixed-Pose Positioning in Smartphone-Based Pedestrian Dead Reckoning Using Hierarchical Clustering

Smartphone-based pedestrian dead reckoning (PDR) systems have gained popularity in indoor positioning due to their low cost, minimal preliminary work, and reduced environmental dependence. However, the performance of PDR systems may be weakened by various poses, in which smartphones are carried. Existing methods generally involve a limited number of poses and adopt a pose classification-based strategy which adjusts positioning methods according to classification results. However, the complexity and variability of daily life poses make classification difficult, thereby affecting positioning performance. In this article, we propose a hierarchical clustering-based algorithm which enables mixed-pose positioning without the need for complex classification algorithms. Our approach includes common poses as well as additional poses, such as backpack, satchel, shirt pocket, and waist-mounted, covering a wide range of poses. We first cluster the 12 total poses into two and four clusters, respectively, according to step detection and step length estimation, which are two crucial parts of PDR. We then designed step detection and step length estimation strategies for each cluster, respectively. To account for heading offset caused by pose switching, we estimate and compensate for the offset based on going straight/turning motion mode classification results. Experimental results show that our algorithm improves the continuity and accuracy of mixed-pose positioning. The average step length estimation error is within 4% for 25 experimenters, and the average positioning error is within 15 m for a 30-min mixed-pose walking experiment.