Multi-Target Adaptive On-line Tracking based on WIHM

Multi-target tracking based on on-line boosting is a significant technique in computer vision. However, it is very difficult to select the optimal classifier in on-line learning process since tracking often relies on an assumption that the appearance model of target is fixed. This would directly lead to a decline in the performance of on-line boosting. In this paper, we presents a novel on-line multi-target tracking framework based on Weighted Incremental Histogram Model (called WIHM), which can be applied in some static or dynamic scenarios. First, we propose a novel method—WIHM, which is employed to obtain the optimal size of a tracked object. Second, a new update scheme is used to reach local optimum tracking appearance model with high possibility and accuracy. Third, based on the above works, a multi-target tracking framework is proposed to track multi-targets simultaneously. With the appearance model of targets changed continually, our represented approach can track these targets more powerful, especially in dealing with camera motions. Experimental results show the effectiveness and robustness of our method.