基于Sampson误差计算单应的立体匹配  

Stereo Matching Based on Sampson Distance of Computing Homogra phy

作　　者：田谨思[1] 苏剑波[1] 

机构地区：[1]上海交通大学自动化系,上海200030

出　　处：《计算机工程与应用》2005年第3期7-9,30,共4页Computer Engineering and Applications

基　　金：国家自然科学基金项目(编号:60275042);上海市教委曙光计划项目资助

摘　　要：立体匹配是立体视觉重要的研究内容,文章在摄像机无标定的前提下采用Sampson误差估计立体图像对的平面单应,由单应计算基础矩阵,进行图像特征点匹配。先用Forstner算子提取角点,按照灰度差相似性准则进行初始匹配,然后在Sampson误差模型下求解代价函数的最优解,解决单应矩阵元素方程组超定问题。在平面单应及对极几何约束下进行图像特征点匹配,可获得射影意义下的象点重构。实验结果表明该方法能准确快速地匹配特征点。Stereo matching is an important research area in stereo vision.This paper uses feature points of uncalibrated image pair to compute the homography matrix based on Sampson error model.Homography matrix is used to compute the fundamental matrix and search the corresponding point elements of the object's structure.The corners are detected by Forstner algorithm and the original matching is based on the gray similarity criterion.The optimal cost function is established under the Sampson error estimation model,so the overdetermined equations are resolved.The corners are matched under the epipolar constraint and homography constraint.And the structure is reconstructed under projective geometry model.Experimental results show that the method is accurate and fast.

关 键 词：特征点匹配 单应 Sampson误差 重构 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]