一种改进的快速相关跟踪算法

Improved fast correlation tracking algorithm

在最大近邻距离(MaximumCloseDistance,MCD)相关跟踪算法的基础上进行改进,提出了最小远离距离(MaximumFarDistance,MFD)算法。该算法通过单调降低累加和的上限,快速终止非匹配点处的计算,最终减小了总体计算量。为应付MCD算法中存在的模板漂移问题,对MFD算法采用的像素相似性阈值进行动态调整,并在出现多个匹配点的情况下用另外的相关算法做最终筛选,从而在保持了MCD算法抗噪声、抗局部遮挡等优点的同时,显著抑制了模板的漂移。算法还将相关匹配与伺服系统的控制特性结合起来进行考虑,更加合理地选择匹配区域,从而允许更快的伺服控制速度。

A correlation tracking algorithm named Minimum Far Distance（MFD） was proposed to improve the existing Maximum Close Distance （MCD） algorithm. By monotonically decreasing the upper limit of summation, MFD can quickly terminate accumulating on the non-matching points, and finally reduce the total calculation. In order to deal with the template drift problem of MCD, MFD adjusts the pixel close threshold dynamically, and uses another correlation method to filter from multiple matching points that may be exist. As a result, MFD suppresses template drift remarkably, while preserves the virtues of MCD such as standing against noise and partial occlusion. The algorithm also combines correlation matching with the control characteristic of servo system, selects the matching area more reasonably, and thus enables a faster servo control speed.