结合Kalman滤波器的Mean-Shift跟踪算法  

Mean-Shift tracking algorithm combined with Kalman filter

作　　者：刘继艳[1] 潘建寿[1] 吴亚鹏[1] 王宾[1] 付勇[2] 

机构地区：[1]西北大学信息科学与技术学院,西安710127 [2]西北工业大学电子信息学院,西安710072

出　　处：《计算机工程与应用》2009年第12期184-186,197,共4页Computer Engineering and Applications

基　　金：陕西省自然科学基金No.F0306~~

摘　　要：针对经典Mean-Shift算法要求相邻两帧间目标模板区域必须重叠的缺陷,结合Kalman滤波器,提出了改进算法。算法首先将Kalman滤波器预测的目标位置作为Mean-Shift算法中的初始搜索中心进行跟踪,然后再将Mean-Shift算法得到的新的目标位置作为下一帧Kalman滤波器的输入参数,循环执行。实验证明,该算法能够解决由于目标运动速度突然变化以及目标快速运动情况下所带来的相邻两帧间目标模板区域非重叠问题,而且对于一般的遮挡问题也能得到较好的效果。Classical Mean-Shifi algorithm requires that the adjacent two frames target template must have overlap areas.In order to resolve this weak point,an improved algorithm combined with Kalman filter is proposed.First,Kalman filter predicts a position which is to be the initial search center in Mean-Shift algorithm to track.Then,Mean-Shift gets a new target position which is used to be the input parameter of the next Kalman filter.Experimental results show that the algorithm not only can resolve problem which the adjacent two frames target template areas do not overlap because of target velocity changing suddenly or target moving lastly,but also has good performance in ordinary shelter problem.

关 键 词：MEAN-SHIFT 快速运动目标跟踪 KALMAN滤波 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]