Efficient mean shift tracking via particle swarm optimization for multi-articulated human body features

Easily falling into local extremum, plateaus, and fast moving targets could't tracked, which are main handicap to mean shift application, especially in those cases to track the multi-articulated human body fine features. Based on the analysis of the causes of the mean shift, particle swarm optimization is introduced into the mean shift to solve this problem in this paper. Here, the mode estimation is cast as a problem of goal seeking for the particle swarm while it moves through the image data space. Local extremum and plateaus can be avoided through information exchange between each particle of the swarm, and the target candidate positions are added, thereby converging at the mode values of the target candidate region efficiently. At the same time, the tracking ability is ameliorated even if having the occlusion, rapid movement. Experimental results of tracking on several image sequences demonstrate the proposed algorithm is robust and improve the tracking veracity of the multi-articulated human body fine features moving objects.