核函数带宽自适应的Mean shift目标跟踪算法

针对Mean shift核函数带宽不能实时改变的缺陷，提出一种仿射变换下的核函数带宽可变的Mean shift跟踪算法。以仿射变换来描述目标尺寸随时间的变化，分别利用Mean shift和连续两帧中匹配窗口的最大相关系数来得到仿射变换的参数。以比例参数计算出带宽的实时变化，其它参数可提供Mean shift更好的起始位置。这些工作改善了Mean shift算法在目标尺寸变化时的跟踪效果。实验证明，本算法能够有效地跟踪尺寸变化的目标，并且具有更好的实时性。

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

A new algorithm of variable bandwidth target tracking based on Mean shift is put forward to improve deficiency that the bandwidth of kernel function of Mean shift is not changeable. The variation of target with time is described by affine transform. Mean shift method and the method based on the biggest correlation coefficient between two match windows in two continuous frames are respectively used to get the parameters of affine transform. The change of bandwidth in time is calculated with scale parameter of affine parameters, and other parameters provide a better initial position of tracking. These works improve the effect of Mean shift when size of target changes in real time. The algorithm is proved to have better effect and satisfy the real-time request through experiments.