一种基于协方差估计的均值偏移对象跟踪算法

A mean shift object tracking algorithm based on covariance estimation

针对现有运动目标跟踪算法对目标大小、形状变化的适应能力较差,且不能对目标的旋转进行跟踪的问题,提出一种改进的目标跟踪算法。该算法是均值偏移算法的进一步扩展和延伸,在估计目标位置的同时用协方差矩阵来描述目标形状,结合色彩直方图,处理对象的角度和形状、大小发生变化时的跟踪问题。实验结果表明:改进的算法在不同环境下跟踪目标的鲁棒性很好,极大地提高了跟踪精度,具有很强的实用性。

Considering the problems that the current tracking algorithms always fail in the adaption of object’s changing in size and shape and cannot follow when the object rotating,an improved object tracking algorithm was proposed which was an extension of the mean shift procedure.The new algorithm utilizes a covariance matrix to describe the object’s shape while estimating its position,combines with the color histogram,and processes the tracking problem while the target object keeps changing in angle,size or shape.The experiment shows that algorithm can robustly work in different environment and greatly raise the tracking accuracy.