跨摄像头运动目标检测与识别  

Moving Targets Detection and Recognition Across Multiple Widely Separated Cameras

作　　者：梅江元[1] 司玉林[1] 高会军[1] 

机构地区：[1]哈尔滨工业大学智能控制与系统研究所,黑龙江哈尔滨150080

出　　处：《自动化技术与应用》2011年第11期43-46,共4页Techniques of Automation and Applications

摘　　要：跨摄像头运动目标跟踪是目前智能网络监控的一个重点研究方向。本文首先提出了一种局部背景更新法,有效地解决了背景更新的稳定性与平滑性等问题;其次,本文提出了一种基于自适应阈值的运动目标提取方法,同时通过形态学滤波有效地去除了提取目标时的噪声干扰与空洞问题;最后,文章提出了一种基于YCbCr通道的运动目标直方图特征提取算法,并在特征匹配时加入了平移因子,在测试集合中得到了非常好的匹配性能。Multi-camera tracking is a key research direction of intelligent network monitoring. Firstly, this paper proposes a method based on local background updating, it effectively solves the problems of time-delay and instability. Secondly, a moving object extraction method with adaptive threshold is put forward in this work, and at the same time, morphological filtering is introduced into this algorithm to remove the disturbance and cavities in target extraction. Lastly, the work presents one kind of target matching strategy with statistical histogram, and it gives a very good matching performance.

关 键 词：跨摄像头 运动目标检测 直方图匹配 运动目标识别 

分 类 号：TN919.7[电子电信—通信与信息系统] TP391.4[电子电信—信息与通信工程]