基于二次加权Mean-Shift的目标实时跟踪

Real-time object tracking based on Mean-Shift with dual-weight

针对经典Mean-Shift跟踪算法需要多次迭代才能达到收敛的缺点,提出一种高效的Mean-Shift跟踪算法。在使用颜色空间作为目标特征的跟踪系统中,目标本身往往可以表征为区别于背景的颜色特征,而颜色特征的分布则与偏移向量的权值相对应。通过分析跟踪算法中不同的权值对收敛速度的影响,对加权系数进行了二次加权,使改进的算法只需要一次粗定位和一次精确定位2次迭代便可准确地对目标进行定位。试验结果表明,该算法在保证了经典算法准确性的同时,大大加快了向目标收敛的速度。

An improved Mean-Shift trailing algorithm is proposed to resolve the short-coming of classical Mean-Shift trailing algorithm which needs multi-iterations to converge.In the object tracking systems,the color space is often adopted as the feature space,and the objects usually have unique colors so as to distinguish it selves from the background.In Mean-Shift algorithm the color feature＇s distributions are associated with the weights of the Mean-Shift vector.Through analyzing influences of different weights may have on the converge speeds,we reweight the weight coefficients,and permit the algorithm to accurately locate the target only through two iterations： a coarse one and a fine one.Experimental results demonstrate that the presented method is more efficient than the classical Mean-Shift algorithm,while it accelerates the speed of target convergence.