基于改进的Mean-Shift算法的人体跟踪  

People Tracking Based on Improved Mean-Shift Algorithm

作　　者：郑晓峰[1] 许增朴[1] 于德敏[1] 王永强[1] 

机构地区：[1]天津科技大学机械工程学院

出　　处：《微计算机信息》2009年第31期93-95,共3页Control & Automation

摘　　要：针对Mean-Shift算法在人体进行变加速随机运动和发生遮挡时容易跟踪失败的问题,提出了扩展卡尔曼滤波器和Mean-Shift结合的人体跟踪算法。利用扩展卡尔曼滤波器对人体运动参数进行估计,然后再利用Mean-Shift算法得到人体的位置跟踪。通过对当前窗口Bhattacharyya系数的计算来判断人体是否被遮挡,当人体被遮挡后,关闭扩展卡尔曼滤波器,对人体位置进行线性预测。试验表明,该算法较好地解决了人体在变加速随机运动以及被遮挡情况下的跟踪问题。An improved Mean-Shift-based people tracking algorithm was proposed to solve the problem, which was poor tracking ability when people did changing accelerated random motion and being occluded. Human motion parameters were estimated by the Extended Kalman Fiher（EKF）, and then Mean-Shift algorithm was utilized to track the people location. The Bhattacharyya coefficient of current window was calculated to determine whether people was occluded, if was, the EKF stopped working and people location was estimated by linear prediction. Experimental resuhs shown that the improved algorithm can commendably track the people in changing accelerated random motion and occulusion.

关 键 词：Mean—Shift 扩展卡尔曼滤波 人体跟踪 BHATTACHARYYA系数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]