重叠视野域内多摄像机目标交换算法  

Multi-camera object handoff algorithm in overlapping fields of view

作　　者：欧阳宁[1,2] 刘钊[1] 

机构地区：[1]电子科技大学电子工程学院,四川成都610054 [2]桂林电子科技大学信息与通信学院,广西桂林541004

出　　处：《激光与红外》2009年第9期995-998,共4页Laser & Infrared

基　　金：桂科基(No.0731020);桂科能(No.063006-5G-4)基金项目资助

摘　　要：多摄像机之间重叠视野区域内目标交接问题是实现多摄像机监控系统持续目标跟踪的关键,传统方法是在摄像机重叠视野区域内采用特征匹配方法或基于视野分界线的目标交接算法。但是特征匹配计算量较大,受摄像机参数设置因素影响较大,实用性不高。而基于视野分界线的目标交接算法由于新目标的检测存在延时,仅利用该方法来实现一致性标号容易出错。近年来,出现了利用投影不变量,计算目标在相邻摄像机中的投影位置来进行目标交接的方法,取得较好效果。但在多个目标相距较近的情况下,投影过程中产生的误差会造成目标对应错误。文中提出一种投影不变量和直方图匹配相结合的改进算法,实验表明能较好解决了上述问题。The implementation of object handoff in overlapping fields of view is one of the key techniques of continual object tracking in multi-camera tracking systems. Feature matching methods and the algorithm based on field of view are adopted in conventional solutions towards this problem. However, FM is not practical in application because it is not computationally efficient and the results heavily depend on the parameter settings of the cameras. The approaches based on FOV suffer from the delay of the detection of newly appeared objects, causing the results of consistent labe- ling are not reliable. Recently,projective invariants method is used to handoff object,but it will make mistakes in the case of multiple objects appear closely. The paper proposes a novel scheme for object handoff based on projective invariants and histogram matching. The experimental results show that our method could solve the problem above.

关 键 词：多摄像机 目标交接 投影不变量 直方图匹配 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]