基于改进Bhattacharyya系数的粒子滤波视觉跟踪算法

Modified Bhattacharyya coefficient for particle filter visual tracking

基于颜色直方图的粒子滤波跟踪通常采用Bhattacharyya系数(B氏系数)衡量目标与候选区域特征模型之间的相似性.分析说明目标内部区域的B氏系数存在大量的峰值,使得粒子滤波跟踪仅能适应目标收缩,无法适应目标的膨胀.为此,提出了一种改进的B氏系数,从理论上分析说明了该系数具有单峰特性,基于该系数的粒子滤波跟踪能同时适应目标收缩和膨胀.分析和实验结果均表明,基于本文提出的改进B氏系数的粒子滤波跟踪对目标快速膨胀和收缩等形变具有较好的鲁棒性和准确性.

Bhattacharyya coefficient（BC） is always utilized as the measurement of similarity between models of object and candidate region in particle filter visual tracking based on color histogram. This paper analyzes ＇and explains that there exists thousands of peaks of BC of internal regions in object such that particle filter tracking with BC only adapts to object shrinking while not to object dilating. Therefore, a modified version of BC is proposed, which is proved to be single-peak, and the particle filter based on which is applicable for both object shrinking and dilating. Both analysis and experimental results demonstrate the robustness and accuracy of particle filter tracking based on the modified BC.