无重叠视域多摄像机环境下的目标匹配算法  

Objects matching algorithm of multi-camera across non-overlapping views

作　　者：吴从中[1] 张腾飞[1] 彭伟[1] 陈玮 

机构地区：[1]合肥工业大学计算机与信息学院,合肥230009 [2]山东科技大学信息与电气工程学院,济南250022

出　　处：《计算机应用》2010年第12期60-62,共3页journal of Computer Applications

摘　　要：大区域的视频监控会用到多台无重叠视域的摄像机,目标在时间和空间上都是分离的,如何将不同摄像机中相同目标进行匹配是无重叠视域多摄像机目标跟踪的关键问题。提出一种基于时空和表现模型关系的算法来实现目标匹配,利用核密度估计方法计算时空变量多元概率密度来掌握交互摄像机之间的关系,通过摄像机之间的亮度转移函数(BTF)来掌握目标在摄像机之间表现模型的改变。采用真实视频和仿真方法对算法进行实验,实验结果表明此算法匹配效果良好。Multiple non-overlapping cameras are required in visual surveillance of wide areas,and the observations of objects are often widely separated in time and space.The key problem of multiple-camera objects tracking is how to match the same object in different cameras.A objects matching algorithm based on space-time and appearance model relationship was proposed in this paper.The algorithm learned the inter-camera relationship by kernel density estimation to count multivariate probability density of space-time variables,and used Brightness Transfer Function（BTF） from one camera to another to handle the appearance model change of an object as it moves from one camera to another.Experiments with real videos and simulation validate the proposed approach.

关 键 词：视频监控 无重叠视域多摄像机 目标跟踪 亮度转移函数 目标匹配 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术] TP391.41[自动化与计算机技术—计算机科学与技术]