3D Human Motion Tracking Combining Part Segmentation and Particle Filter

In spite of the fact that the stochastic sampling decreases the risk to be trapped in a local extremum,the global optimum can not be guaranteed in the stochastic sampling methods.In this paper,we propose a human motion tracking system embedding body part segmentation in a particle filter framework.Our system uses Condensation algorithm to propagate and evaluate particles,which are then iteratively updated with expectation maximization based body part segmentation algorithm.During the iteration,the body pose information derived from particles is used for partial segmentation.Then the particle distribution is specified by the segmentation results.This strategy steers the particle distribution towards regions with a high likelihood and increases the chance to find the global optimum with fewer particles.The experimental results on HumanEva-II dataset illustrate the effectiveness of our approach.The performance of our proposed method is better than that of Condensation algorithm and annealing particle filter.