基于边界力的Mean-Shift 核窗宽自适应算法

针对Mean-Shift(中值漂移)算法中核函数带宽不能实时改变的缺陷,提出一种基于边界力的Mean-Shift核函数带宽自适应更新算法.在分析目标加权核直方图模型的基础上,引入区域似然度以提取目标的局部信息.然后,比较相邻帧间的区域似然度并构建边界力.通过对边界力的计算,得到边界点的位置,进而自适应地更新核函数带宽.实验结果表明,这些工作改善了Mean-Shift 算法在目标尺度和形态发生变化时的跟踪效果,并且可以满足实时性的需要.

Algorithm of Adaptive Kernel-Bandwidth for Mean-Shift Based on Boundary Force

An adaptive scale updating algorithm based on boundary force is presented to improve the deficiency that the kernel-bandwidth of Mean-Shift is not changeable. Based on the analysis of weighted histogram of the target feature, this paper introduces a region likelihood to extract local information of the target. Then, by comparing the region likelihood in successive frames, it constructs a boundary force to locate the boundary points of the target model and updates the bandwidth of kernel-function adaptively. The experimental results show that the proposed method improves the effect of Mean-Shift when the size or shape of target changes and satisfies the real-time request.