3D Human Motion Tracking Using Progressive Particle Filter

Abstract

The work proposes a model-based 3D human motion tracking algorithm, Progressive particle filter, with a single unconstrained camera. Particle filter is an useful algorithm for human motion tracking, but it suffers from the degeneracy problem and huge computation. The study improves the sampling efficiency by integrating the mean shift trackers into each particle toward each local maximum for raising the accuracy. Besides, we also combine the hierarchical searching approach to decompose the high dimensional space into three low dimensional spaces for reducing the computational cost. Experimental results show the proposed algorithm can successfully reduce the computational cost and track more accuracy than classical particle filter.