带宽自适应的Mean-Shift跟踪算法研究  

Study of bandwidth adaptive Mean-Shift tracking

作　　者：罗镇宝[1] 张永科[1] 吴钟建[1] 

机构地区：[1]西南技术物理研究所电子工程部,成都610041

出　　处：《计算机应用》2009年第6期1683-1686,共4页journal of Computer Applications

摘　　要：针对图像跟踪中目标的尺度和旋转变化,将Lindeberg的尺度理论与Mean-Shift算法结合起来,提出了一种带宽自适应Mean-Shift跟踪算法。该算法在Mean-Shift的框架下,将尺度和旋转量与平移量同等看待,通过求解核函数带宽,计算出目标的变化参数,最终精确定位目标。另外,引入SAD算法对目标进行先期粗略定位,克服了目标做无规律大位移运动时Mean-Shift算法跟踪效果不佳的问题,同时也降低了Mean-Shift算法的迭代收敛次数。大量实验仿真表明,该算法对目标的仿射变化、非刚性形态变化,以及无规律的大位移运动具有有效性和鲁棒性。Aiming at the target scale and rotation motions in the tracking image, a bandwidth adaptive Mean-Shift algorithm combined with the Lindeberg scale theory was proposed. Based on the Mean-Shift mode, the proposed algorithm tracker regarded the target＇s scale and rotation as the equal variables with displacements, and the parameters of the target motions were obtained to locate target precisely in the image by computing the kernel bandwidth. Moreover, the simple location of target firstly through introducing Sum of Absolute Differences （SAD） into the Mean-Shift tracker can solve the problem that the tracker does not handle large erratic displacements effectively and can also reduce the Mean-Shift iterative times. Simulation Experiment results show that the proposed tracking algorithm is robust and effective to target affine motions, non- rigid shape changes and large erratic displacement.

关 键 词：带宽自适应 Mean—Shift算法 目标跟踪 绝对差和 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]