一种针对视频对象快速移动和遮挡的改进mean-shift跟踪算法  

A robust mean-shift tracking method under fast-moving and occlusion

作　　者：唐勇[1] 孙磊[1] 孙宾[1] 

机构地区：[1]燕山大学信息科学与工程学院,河北秦皇岛066004

出　　处：《燕山大学学报》2010年第1期52-55,66,共5页Journal of Yanshan University

基　　金：国家自然科学基金资助项目(50675189)

摘　　要：针对经典mean-shift算法在目标发生快速移动或受到局部遮挡时容易跟踪失败的问题,提出一种改进的mean-shift算法。首先以Bhattacharyya系数作为目标运动状态判断因子:当目标发生快速移动时,采用SSD(Sumof Square Difference)算法进行全局搜索,以此来指定当前帧mean-shift算法的初始跟踪位置;当目标受到局部遮挡时,采用背景相似度算法,对目标模板和候选模型区域中像素点进行加权,降低背景像素的干扰,然后由SSD算法进行局部匹配,对mean-shift算法的跟踪结果进行校正。经实验表明,改进算法对发生快速移动或受到局部遮挡的目标,能进行有效的跟踪。An improved mean-shift-based tracking algorithm was proposed to solve the poor tracking ability under fast-moving and partial occlusion. Bhattacharyya coefficient analyzing was used to determine the tracking status： when the object moves fast, SSD algorithm is applied to choose the start point for the mean-shift tracking; when the object undergoes partial occlusion, firstly the background pdf is used to separate the object region from the background; secondly, SSD algorithm is employed to modify the result from the mean-shift tracking. Experiment results of video sequences demonstrate that the proposed method can track the objects stably and accurately for fast-moving and occlusion problems.

关 键 词：目标跟踪 mean—shift 快速移动 遮挡 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]