Multi-Camera Collaboration Based on Trifocal Tensor Transfer

Abstract

针对复杂遮挡环境下多摄像机协同的问题,提出一种基于三焦点张量点转移的多摄像机协同目标定位方法.该方法利用单摄像机视图中头部检测的结果进行对极匹配,然后利用虚拟顶视图和摄像机视图之间的三焦点张量关系进行点转移,确定行人在虚拟顶视图中的位置.该方法的优势在于:不需要标定摄像机参数,不需要假设物体在共同平面上运动,不存在利用对极关系进行点转移时出现的交点退化的问题.与传统方法的实验结果对比表明了该方法的有效性和准确性.;This paper presents a head detection and trifocal tensor pointer transfer based multi-camera collaboration. In this approach, people’s head position is detected after background subtraction and tracked by Kalman and PDA. Trifocal tensor transfer is used to locate objects in the virtual top view by the corresponding head points in two camera views. Compared with existing approach, the approach doesn’t need camera calibration and coplanar precondition, nor does the deterioration of epipolar transfer exist. The experimental comparison between this approach and other traditional ones proves its effectiveness and accuracy.