均值漂移目标跟踪的两级窗宽更新算法

目标尺度发生较大变化时，固定窗宽的均值漂移(Mean shift)目标跟踪算法不能进行有效跟踪。为此，提出一种两级更新的自适应窗宽计算方法。采用Mean shift跟踪算法对目标中心进行初次定位，并确定窗宽的大小，设置感兴趣区域，结合已建立的背景模型，在感兴趣区域内利用背景减除技术二次确定目标的中心及窗宽大小，通过比较2次目标区域与目标模型之间的Bhattacharyya系数，选择系数较大的区域作为最终跟踪窗口。实验结果表明，该方法能够对尺度变化明显的运动目标自适应确定跟踪窗宽，并减小传统Mean shift跟踪方法背景目标颜色对目标特征提取的影响。

Two-step Window Bandwidth Update Algorithm of Mean Shift Object Tracking

The target tracking algorithm of Mean shift with fixed window bandwidth can not trace an object effectively when the moving object size changes.To solve this problem,a two-step update algorithm of adaptive window bandwidth is proposed.The Mean shift algorithm is used to compute the center and the size of target,and then they are computed again by the algorithm based on background subtraction method in the Region of Interest(RoI).By comparing the two Bhattacharyya coefficients between two target regions with the target model,final tracking window is obtained by selecting the region which has larger Bhattacharyya coefficient.Experimental results show that the algorithm can adaptively confirm the tracking window when the size of target changes greatly,and decrease background color effect on the target feature extraction.