Robust object tracking using kernel-based weighted fragments
Guanbin Li, Hefeng Wu · 2011
In this paper we propose a novel kernel-based tracking approach using weighted fragments. We represent the target with multiple fragments and define the weight of each fragment using the proportion of object and background distributions. We invoke an independent mean shift tracker for each fragment and then combine the tracking results of all the fragments in a linear weighting scheme. The proposed algorithm is computationally efficient enough to be executed in real time. Experimental results verify that the proposed algorithm better handles the problems of partial occlusions and pose changes.