基于灰色预测和Mean-Shift的抗遮挡跟踪算法  

Anti-occluding Tracking Algorithm Based on Grey Prediction and Mean-Shift

作　　者：王长有[1,2] 刘皓[1,3] 张海强[1,3] 胡敏[1,3] 李湘伟[1,3] 

机构地区：[1]重庆邮电大学工业物联网与网络化控制教育部重点实验室,重庆400065 [2]重庆邮电大学应用数学研究所,重庆400065 [3]重庆邮电大学自动化学院,重庆400065

出　　处：《控制工程》2017年第7期1323-1328,共6页Control Engineering of China

基　　金：国家自然科学基金(11372366);重庆市自然科学基金(cstc2012 jjA20016)

摘　　要：针对经典的Mean-Shift算法很难准确跟踪有遮挡的运动目标这一问题,提出了一种改进的运动目标跟踪算法。跟踪中利用Bhattacharyya系数判断跟踪目标是否出现了遮挡,若出现遮挡,则通过利用改进的GM(1,1)预测模型,根据历史目标位置信息对后续帧中目标位置进行预测,若目标脱离遮挡,则继续使用Mean-Shift进行后续跟踪。通过对不同环境下的视频序列进行测试,其结果表明算法能够对发生遮挡后的目标进行持续、稳健的跟踪。The Mean-Shift algorithm fails to track objects easily while it is occluded, so a new objecttracking method is proposed. This method combines the Mean-Shift algorithm and the GM （1, 1） model withoptimized background values. The Bhattacharyya coefficient is used to determine whether the target isoccluded in the process of tracking. When the object is occluded, the target position in the follow-up frame ispredicted by the GM （1,1） prediction model and the historical target location information. If the target is outof occlusion, the Mean-Shift algorithm is continued to be used to conduct the follow-up tracking. Experimentresults of variant environment video sequences demonstrate that the proposed method can track the objectsstably and accurately during occlusions.

关 键 词：Mean—Shift算法 灰色预测 BHATTACHARYYA系数 遮挡 目标跟踪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]