Contour cue based particle filter for monocular human motion tracking

Particle filter is widely used in human motion tracking but its efficiency is low. A contour cue based particle filter algorithm is proposed in this paper for the human motion tracking in a mark-erless monocular video. The likelihood of sampled particles is measured by chamfer distance between two contours. One con-tour is extracted from the video image. The other is transformed from the sampled particle. The value of likelihood is the weight of corresponding particle. Then weighted particles are optimized by Levenberg-Marquardt method to make the final estimation closer to the posterior distribution of motion state. Apart from this, the skin part of human body is detected to constrain the sampled particles when the contour feature points are not sufficient with large occlusion. The experiment result shows that the contour cue based method is more efficient than the edge method.