基于边界沙包核函数的Mean-Shift跟踪算法  

Mean-Shift tracking algorithm based on boundary detection and sandbags kernel function

作　　者：曹义亲[1] 肖金胜[1] 黄晓生[1] 

机构地区：[1]华东交通大学软件学院,南昌330013

出　　处：《计算机应用研究》2015年第11期3475-3479,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61365008);江西省自然科学基金资助项目(20142BAB207025);国家教育部人文社科资助项目

摘　　要：针对传统的Mean-Shift跟踪算法,使用单个颜色特征定位目标易受相似目标与背景的干扰导致跟踪失败以及跟踪窗口尺寸不能自适应跟踪目标变化的问题,提出一种基于颜色特征与边界特征相融合的目标表示方法和沙包核函数Mean-Shift尺寸自适应算法。在跟踪中,颜色特征和边界特征根据各个特征的可靠性进行实时性更新;同时,在跟踪窗口中心和边界定位的基础上,由候选目标跟踪窗和分块目标跟踪窗的边界距离变化对核窗宽大小进行更新。实验结果表明,该算法目标定位的精确性更高,在目标尺寸增大和减小的情况下,平均每帧耗时比传统的基于矩形窗和椭圆窗自适应跟踪算法更少,提高了跟踪性能,满足实时性要求。The classical Mean-Shift algorithm is only used single color histogram represents the target which easily leads to track failure, especially if the scene contains other objects characterized by a color distribution similar to that of the object of interest. The other hand, window function can not be adaptive in the object tracking. So the paper proposed a new algorithm for tracking target that combined color histogram with boudary describing goal ,wihch was the real-time update according to the reliability characteristics. As well as it proposed a newly adaptive window of sandbags which was adaptive based on the size change between the candidate and the block target. Take the experimental result into consideration, the paper proposed new algorithm＇ s had well accuracy in target location. The new algorithm＇ s average per frame time-consuming is more decreased than the traditional adaptive tracking algorithm based on a rectangular window and elliptic window when the goal scale getts bigger or smaller. It improves the tracking performance, meets the real-tlmer equirements.

关 键 词：MEAN-SHIFT 目标跟踪 颜色直方图 沙包窗核函数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP301.6[自动化与计算机技术—计算机科学与技术]