运动像机下基于粒子滤波的多目标跟踪算法  

Multi-target tracking algorithm of moving camera video based on particle filter

作　　者：陆意骏[1] 陈一民[1] 黄诗华[1] 

机构地区：[1]上海大学计算机工程与科学学院,上海200072

出　　处：《计算机工程与设计》2010年第7期1513-1515,1569,共4页Computer Engineering and Design

基　　金：国家科技支撑计划课题基金项目(2006BAK13B10);上海市重点学科建设基金项目(J50103)

摘　　要：针对摄像机运动情况下的多目标跟踪问题,提出了基于粒子滤波的跟踪算法。在粒子滤波算法基础上,将二阶自回归过程作为系统状态转移模型,HSV颜色直方图作为观测模型,对视频中多个目标的位置、大小进行跟踪。实验结果表明,该算法能实时正确地跟踪多个目标,并对局部遮挡有较好的鲁棒性,也能在目标短暂消失导致跟踪失败后,在目标重新出现后及时捕获并继续进行跟踪。A tracking algorithm based on particle filter is proposed to deal with multi-target tracking in a camera motion situation.In the proposed algorithm,the position and scale of targets in the video are tracked by using a second-order auto-regressive process as the tran-sition model and an HSV histogram as the observation model.Experimental results show that the proposed algorithm can track multi-target correctly in real-time and is robust to the partial occlusion.In addition,it can recover from tracking failure due to the target out of camera for a while.

关 键 词：多目标跟踪 摄像机运动 粒子滤波 自回归过程 HSV颜色直方图 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]