带宽自适应的Mean-Shift跟踪算法

Algorithm of Target Tracking Based on Mean-Shift with Adaptive Bandwidth

针对传统Mean-Shift目标跟踪算法在进行物体跟踪时核函数带宽不能实时更新的问题,提出了一种基于比较中心加权直方图与边缘加权直方图的巴氏相似度的带宽自适应算法。首先,手动选取需要跟踪的目标,计算目标模板的中心加权直方图与边缘加权直方图以及二者的巴氏相似度HistDist1;然后,在当前帧通过Mean-Shift迭代找到目标的中心,并计算出候选模板的边缘加权直方图以及目标模板中心加权直方图与候选模板边缘加权直方图的巴氏相似度HistDist2;最后,在设定范围内比较巴氏相似度HistDist1与HistDist2的大小,得到目标的尺寸变化情况。试验结果表明,该算法可以适应尺寸发生变化的目标跟踪,实现了核函数带宽的自适应。

In allusion to the issue that the kernel functional bandwidth can not be updated in real time in traditional Mean-Shift target tracking algorithm,a new band-width-adaptive algorithm based on Bhattacharyya similarity between center weighted histogram and edge weighted histogram is proposed.First,the required target is selected manually in the first frame.Bhattacharyya similarity HistDist1between target template center weighted histogram and edge weighted histogram is calculated by using the Bhattacharyya.Then,the center of target in the current frame is searched by Mean-Shift algorithm.The Bhattacharyya similarity HistDist2between target template center weighted histogram and candidate template edge weighted histogram is calculated as similar to the HistDist2.Finally,the Bhattacharyya similarity HistDist2 is compared with HistDist1in a set range,the variation of target size is obtained.The result shows that the algorithm proposed in this paper can be adaptive to the scale-change of target,therefore the self-adaption of the kernel functional bandwidth is implemented.