基于线性预测和加权投票的分块Mean-Shift行人跟踪算法  

Pedestrian Tracking Algorithm Baded on Mean-shiftand Linear Quadratic Estimation

作　　者：周妍[1] 葛万成[1] 

机构地区：[1]同济大学电子与信息工程学院,上海200092

出　　处：《信息通信》2013年第10期43-45,共3页Information & Communications

基　　金：德国罗德与施瓦茨公司资助

摘　　要：行人目标跟踪对公共交通规划、服务改善以及对大客流量的预测具有极重要的意义。而广泛采用的Mean-Shift算法及其改进算法在抗遮挡性能方面仍有非常大的改进空间。对此,文章针对行人轨迹交错或发生互相遮挡时可能引起的漏检及错检问题,结合基于分块的色调直方图、线性预测,提出了一种自反馈的混合检测方法。测试结果表明,该方法在行人发生遮挡或行径路线交叠时能够一定程度上减少错跟和漏跟,从而提高算法的鲁棒性。Pedestrian tracking performs a vital role in the transportation planmng and large passenger now prealcnon. Althougn conventional Mean-Shift and its improved methods are introduced, miss tracking or error tracking always happens when pedes- trians has＇ frequent occlusions. This paper presents a track algorism based on local hue histogram and linear quadratic estimation to predict the possible location of the pedestrian and do readjustment with the actual location of the pedestrian. Results shows that the tracking algorithm has a better performance in case that pedestrians has frequent occlusions and reduce the possibility of miss tracking or error tracking.

关 键 词：行人跟踪 Mean-Shift法 线性预测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]