Articulated 3D human pose estimation with Particle Filter based Particle Swarm Optimization

Abstract

We propose a new Particle Filter (PF) based Particle Swarm Optimization (PSO) algorithm for 3D articulated human pose estimation. The sampling covariance and annealing factor items are incorporated into the velocity updating equation of PSO, which are initiated with appropriate values at the beginning of PSO iteration, and decreasing (`annealed') by reasonable steps. The new algorithm can, in some degree, mitigate the not sufficiently reliable image likelihood problem. Experimental results on HumanEvaI data set show that compared with annealed particle filter and standard particle filter, the proposed algorithm can achieve lower estimation errors in tracking real-world 3D human motion.