Abstract
Markerless human body part tracking and pose estimation have recently attracted intensive attention because of their wide applications. The vision-based approaches to solve the problems of motion parameters capturing always meet two challenges. 1) how to solve the parameter estimation problem in high-dimensional space, and 2) how to deal with the missing observation information due to occlusion. To solve the two problems, we proposed a vision-based method combining the Annealed Particle Filter (APF) [6] with a pre-trained correlation map and temporal constraint. This paper proposes a system for capturing motion parameters of walking human object in indoors and outdoors. To solve the problem with shadow when we track the walking people in outdoors, we use the HSV model to remove the shadow. Compare to the traditional APF [6], our method needs less operation time and has more accurate result. Because of the pre-trained correlation map and temporal constraint, our method also has better performance than the tradition APF when self-occlusion occurs.