2多摄像机环境下的约束关系.doc

_________________________ 基金项目:国家自然科学基金项目 收稿日期: 改回日期: 第一作者简介:陶霖密(1962 - )，男，副教授，博士（2001年清华大学），主要研究领域为计算机视觉、人机交互。E-mail：tao.linmi@ 摄像机几何约束及人体定位 陶霖密 于亚鹏 邸慧军 孙洛 （清华大学计算机系，北京100084） 摘要 人体定位一直是计算机视觉领域的热点课题之一，是人机交互的基础。本文提出了多摄像机条件下的投影关系的两类几何约束，并提出了基于几何约束的人体定位方法。人体定位算法的精度计算依赖于实验中人体在世界坐标系中的精确位置的获取，而这一直是计算机视觉实验中的难点。本文提出了基于虚拟数据的新的人体定位实验方法，可以方便准确地获取到人体在世界坐标系的位置。由于几何约束的完备性，定位过程中无需对摄像机的位置和姿态进行假设，也不??要限制人体在场景中的位置（遮挡关系）、行为等。采用虚拟数据与真实数据相结合，可以更准确的验证方法的精确度和适应性。实验结果表明，本文的算法运行稳定且定位精度高。在定位结果的基础上，本文进一步给出了人体在世界坐标系中的运动轨迹。 关键词 几何约束 投影变换 EM算法 虚拟实验 中图法分类号：TP301.6 文献标识码：A 文章编号：1006-8961(2011 ) - - Geometric Constraints of Cameras And People Locating Tao Linmi, Yu Yapeng, Di Huijun, Sun Luo (College of Computer Science and Technology, Tsinghua University, Beijing, 100084) Abstract The human body locating has been one of the most popular subjects in computer vision, which is the basis of the human-computer interaction. In this paper, we propose two types of geometrical constraints based on multi-camera and a people locating algorithm based on them. The accuracy of people locating algorithm relies on the exact location in the world coordinate system got in the experiments, which has been a great difficulty in computer vision experiments. In this paper we present a new people locating experimental method based on virtual data, with which we can more easily and accurately gain the people location in the world coordinate system. Because of the completeness of the geometric constraints, we do not need assume the cameras’ placement and restrict the location (occlusion) or motion of human in the scene. Using virtual and real data in the experiments, it can be more accurate for verifying the accuracy and adaptability of the algorithm. The experimental results show that the algorithm works stably and accurately. In addition based on the results of location, the trajectory for people can be gained. Keywords geometric constraint, homography mapping, EM algorithm,