Abstract: Traditional Mean-Shift based object tracking adopts whole features for tracking，and is hard to track well under object occlusion.A new local feture based method is proposed，which combines the blob modeling and mean-shift together.Firstly，the blob modeling for the tracked object is built，and then each blob is tracked by the Mean-Shift method.Finally the new position of object is determined.The proposed method can select unoccluded blob for object tracking when occlusion occurs.The background-weighted Mean-Shift method is adopted to improve the robustness to the background disturbance.Experimental results show that the method can track the object exactly under the circumstance of partial occlusion，and the performance is better than that of traditional Mean-Shift based method.

Key words: object tracking, partial occlusion, blob, Mean-Shift algorithm

摘要： 传统的基于Mean-Shift的目标跟踪方法利用目标的全局特征进行跟踪，在局部遮挡情况下跟踪效果不佳。提出一种基于团块建模和Mean-Shift相结合的利用目标局部特征的运动目标跟踪方法，对目标进行团块建模，利用Mean-shift算法对各团块进行跟踪，在此基础上确定目标新位置。该方法能够在目标发生局部遮挡时，自动选取未被遮挡的团块的跟踪结果来确定目标的位置。为了提高方法对背景干扰的鲁棒性，采用背景加权的Mean-Shift算法。实验结果表明：该方法在局部遮挡的情况下可较好地进行目标跟踪，跟踪效果优于报导的基于Mean-Shift的方法。

关键词: 目标跟踪, 局部遮挡, 团块, Mean-Shift算法