多摄像机间基于最稳定极值区域的人体跟踪方法  

Human tracking method based on maximally stable extremal regions with multi-cameras

作　　者：张莉[1,2] 刘济林[1] 

机构地区：[1]浙江大学信息与电子工程学系,浙江杭州310027 [2]杭州电子科技大学计算机学院,浙江杭州310018

出　　处：《浙江大学学报（工学版）》2010年第6期1091-1097,共7页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(60534070)

摘　　要：为实现多摄像机视频监控系统中的人体跟踪,提出了一种以最稳定极值区域作为匹配特征的人体跟踪方法.该方法使用图像的灰度信息,在一定程度上减小了摄像机增益及光谱特性对人体跟踪造成的影响.通过把人体跟踪转化为椭圆形区域的匹配,算法将最稳定极值区域拟合成椭圆形区域,在此基础上,选出符合约束条件的候选椭圆形区域,并将其归一化为单位圆形区域,通过在圆形区域内计算旋转不变量、进行直方图密度估计和计算加权平均距离实现椭圆形区域的正确匹配,从而实现多摄像机间的人体跟踪.实验结果表明:该算法可有效实现多摄像机间的人体跟踪.A human tracking method based on maximally stable extremal region（MSER）was established in order to realize the human tracking with multi-cameras in the video surveillance system. The proposed method uses gray information to reduce the difference of the cameras＇ gains and their light spectrum property. The approach transforms the human tracking into elliptic region matching. The method does elliptic region fitting to each maximally stable extremal region（MSER）,and then selects the elliptic regions which meet some constraints. These selected elliptic regions are normalized to unity circular regions by whitening of covariance matrix. The right matched elliptic regions are obtained by rotational invariant vectors calculation,histogram density estimation and weighted average distance calculation. Thus,the human tracking across cameras is realized. Experimental results show that the approach can effectively realize the human tracking with multi-cameras.

关 键 词：人体跟踪 视频监控 最稳定极值区域 区域匹配 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]