核函数带宽自适应的均值漂移目标跟踪算法

Algorithm of video object tracking based on mean shift with adaptive bandwidth of kernel function

针对均值漂移(Mean Shift)算法在跟踪视频目标过程中核函数带宽固定不变的缺陷,提出了一种核函数带宽与目标大小自适应变化的Mean Shift视频目标跟踪算法。用Mean Shift算法搜索到目标,以搜索框中心不变,将搜索窗口扩大,并计算新搜索框的模型及每个像素的核函数权值;通过将每个像素点的核函数权值代替像素值,并利用不变矩计算方法,识别出能框选住目标的椭圆;根据椭圆参数反求新搜索框的大小(核函数带宽)及位置。实验结果表明,该算法能够有效地跟踪大小变化的目标,并且搜索框能较好地与目标大小相适应。

For changing the process of video object tracking in mean shift algorithm is that the bandwidth of kernel function is fixed, a video object tracking algorithm is proposed based on mean shift with adaptive bandwidth consistent with the size of tar- get. Firstly, the video object is found by using mean shift algorithm, and the search box is expanded while its center is fixed. Secondly, the weight of each point and the modeling in the new searching box are computed. The pixel value of each point is re- placed by the weight of each point. And according to invariant moment method, the ellipse which can circle the object is recog- nized. Finally, using the information of the ellipse, the size and location of the new searching box is obtained. The experimental results indicate that the proposed algorithm can track the target efficiently, and the searching box also can be consistent with the size of the object well.