基于特征贡献度的mean shift视频跟踪算法  

Mean shift algorithm for visual tracking based on feature contribution

作　　者：夏瑜[1,2] 吴小俊[1] 冯振华[1] 

机构地区：[1]江南大学物联网工程学院,江苏无锡214122 [2]江苏广播电视大学常熟学院,江苏常熟215500

出　　处：《控制与决策》2012年第7期1021-1026,共6页Control and Decision

基　　金：教育部新世纪优秀人才计划项目(NCET-06-0487);国家自然科学基金项目(60973094;60572034;90820002;61070121);江苏省自然科学基金项目(BK2006081)

摘　　要：特征空间分析是视觉跟踪中的关键问题.针对mean shift跟踪算法中模板匹配问题,提出了特征贡献度概念,有效减少了背景和噪声因素干扰,使重要性特征在匹配中起到关键作用;利用结构二值分布图携带空间结构信息的优点,很好地避免了统计特征的匹配误差,在一定程度上提高了跟踪的精度和鲁棒性.仿真实验结果表明该算法具有一定的优越性和实时性,在跟踪目标存在变形、噪声、遮挡时也可以达到比较理想的跟踪效果.Feature space analysis is the critical factor of visual tracking. Considering the issue of template matching within the mean shift framework, this paper proposes a concept of feature contribution, which can effectively reduce the influence of background feature and noise, and make importance feature play a key role. In addition, the binary distribution of structure introduced can effectively reduce the error of statistical features by spacial information and improve the tracking accuracy and robustness in a certain extent. Experimental results show that the proposed tracking algorithm is effective and also exhibits good result in the presence of noise, deformation and occlusion.

关 键 词：均值漂移 特征贡献度 重要性特征 结构二值分布图 串行融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]