遮挡环境下采用分层局部敏感直方图的实时目标跟踪  

Multilayer locality sensitive histograms based real-time target tracking under occlusion

作　　者：王宣[1,2] 刘晶红[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所,长春130033 [2]中国科学院大学,北京100039

出　　处：《吉林大学学报（工学版）》2015年第2期663-669,共7页Journal of Jilin University:Engineering and Technology Edition

基　　金：'973'国家重点基础研究发展计划项目(2009CB72400105);国防预研基金项目(402070301)

摘　　要：针对局部敏感直方图跟踪算法中缺乏抗遮挡处理机制,在目标被长时间大面积遮挡时易丢失目标的问题,提出了一种基于分层局部敏感直方图特征的实时跟踪算法。该算法将目标模板分为多个顶层区域,分别判断每个顶层区域是否被遮挡,停止被遮挡区域的模板更新。在更新过程中,针对各顶层区域的置信度不同采用不同的更新速度,使环境变化时算法仍保持良好的抗遮挡性能。对视频的测试结果表明:本方法在目标大小为60pixel×60pixel时平均帧速为30帧/s,满足实时性要求。与局部敏感直方图跟踪算法相比,本算法具有更好的抗长时间遮挡和光照变化的性能。The tracking algorithm based on locality sensitive histograms lacks the function of anti-occlusion and fails to track targets stably under long-time occlusion.To overcome this problem,a real-time tracking algorithm using multilayer template based on locality sensitive histograms is proposed.The algorithm first divides the template into a number of top regions;then estimates the occluded state of each top region;and stops the template updating of the occluded top regions.Because the different reliability of each top region,different update levels are taken to maintain the robustness under varied target conditions.The proposed algorithm is tested with two video sequences and results show that the algorithm achieves stable tracking for target size of 60pixel×60pixel.As compared with the locality sensitive histograms tracking,the proposed algorithm can track stably under illumination change and occlusion.

关 键 词：信息处理技术 局部敏感直方图 分层模板 抗遮挡 实时跟踪 

分 类 号：TN911.7[电子电信—通信与信息系统]