核函数带宽自适应的Mean-Shift跟踪算法

为了实现Mean-Shift跟踪算法中的核函数带宽自适应更新,提出基于比较Bhattacharyya系数的新方法。首先用模板中心加权与目标边缘加权的直方图计算巴氏系数,跟踪时用候选目标边缘加权直方图与模板中心加权直方图计算新的巴氏系数,根据两个系数的大小对核带宽进行10%的缩放。实验表明,该方法有效克服了带宽只能缩小的问题,实现了跟踪窗对目标缩放的自适应性。

Algorithm of target tracking based on Mean-Shift with adaptive bandwidth of kernel function

To achieve adaptive bandwidth of kernel function in Mean-Shift tracking algorithm, a new method based on comparing Bhattacharyya coefficients was proposed, in which, the Bhattacharyya coefficient was firstly computed out by using of center weighted and edge weighted histograms of template image, and then, a new Bhattacharyya coefficient during tracking was computed out according to the edge weighted histogram of candidate image and center weighted histogram of template image, finally 10% of the bandwidth of kernel function was expanded or shrunk by comparing two coefficients. The experiment results show that the method can avoid the problem of nonstop shrinking bandwidth effectively, and adapt tracking window to the change of the target size successfully.