多视角多行人目标检测、定位与对应算法  

Detection,location and labeling under a multi-moving-person,multi-view set

作　　者：徐剑[1] 丁晓青[1] 王生进[1] 吴佑寿[1] 

机构地区：[1]清华大学电子工程系,北京100084

出　　处：《清华大学学报（自然科学版）》2009年第8期1139-1143,共5页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(60472002)

摘　　要：为解决多视角视频监控中多行人情况下的遮挡问题,引入三维重建的思想,提出一种基于空间场的多视角多行人检测、定位和对应算法。该算法首先提取各视角运动前景,融合各视角二值前景图像,利用空间场实施重建,然后根据信息融合结果在空间中检测目标并定位,并由空间检测和定位结果确定各视角中目标的对应关系。在3dsM ax合成数据和实际采集数据上进行的实验测试表明,该算法对图像遮挡的处理能力强,计算复杂度低,基本满足实时要求。A multiple people detection, location and labeling algorithm in a multi-moving-person, multi-view set was developed to solve the problems of occlusion. The algorithm bases on a 3 D reconstruction algorithm with a space field model based framework. It extracts moving foregrounds from respective views and fuses them to reconstruct 3-D space field firstly, and then analyses of the space field results lead to detection and location of the humans in 3-D space which are then labelled in all views in the 2-D image space using projection geometry. Tests on real and virtual data show that the framework has robust performance against occlusions, low computational burden, and can be implemented in real time.

关 键 词：多视角 空间场 遮挡 检测 对应 定位 

分 类 号：TP274.5[自动化与计算机技术—检测技术与自动化装置]