Combining Particle Filter with Local Optimization for Human Body Tracking

A new framework combining particle filter with local optimization is proposed to solve the movement tracking of articulated human model.General particle filter does not fit with the high dimension human body tracking,where the dense sampling strategy is impossible.The local optimization algorithm is introduced into the framework of particle filter to solve this problem.First of all,each sample is optimized according to the observation likelihood model；second,the optimized result is used to guide the sampling step at the next time.The experiment shows that our proposed algorithm,using very few samples,can track the three dimensional pose of human body successfully.