基于颜色直方图的EM对象跟踪算法

摘要

基于均值漂移运动目标跟踪的迭代算法,简单可靠,可以方便准确的找到一个基于内核的概率密度函数估计目标的位置.但是该算法对目标尺寸形状变化的适应能力比较差,文章提出一个改进的均值漂移算法.新算法同时估计目标的位置和用协方差矩阵来描述的目标的形状,能够处理对象的角度和形状大小发生变化时的跟踪问题,运用新的算法实现以颜色直方图为基础的非刚性目标跟踪算法.实验表明,该改进的算法在不同环境下跟踪目标的鲁棒性很好,尤其对跟踪目标的形状和尺寸的改变,具有很强的适用性.%The mean-shift algorithm is a simple and robust method for finding the position of a local maximum of a kernel-based estimate of a density function. But this algorithm is incapable of adapting to the target rotation. A new improved mean-shift algorithm is given here that presents a natural extension of the mean-shift procedure, it is the further expansion and extension of the "mean drifT1' algorithm. The new algorithm simultaneously estimates the position of the local mode and the covariance matrix that describes the approximate shape of the local mode. We apply the new method to develop a new color histogram based non-rigid tracking algorithm object. Experiments show that the new algorithm is able to adapt to object rotation and scaling simultaneously and practical,and has a strong suitability.