基于本质矩阵鲁棒估计的球形全景匹配  

Spherical Panorama Matching Based on Robust Estimation of Essential Matrix

作　　者：吴幼丝 邓非[1,3] 万方 WU Yousi;DENG Fei;WANG Fang(School of Geodesy and Geomatics,Wuhan University,Wuhan 430079 ,China;Guangzhou Urban Planning Survey and Design Institute, Guangzhou 510060, China;Collaborative Innovation Center for Territorial Sovereignty and Maritime Rights,Wuhan University,Wuhan 430079, China)

机构地区：[1]武汉大学测绘学院,湖北武汉430079 [2]广州市城市规划勘测设计研究院,广东广州510060 [3]武汉大学国家领土主权与海洋权益协同创新中心,湖北武汉430079

出　　处：《测绘地理信息》2018年第3期119-123,共5页Journal of Geomatics

基　　金：国家自然科学基金资助项目(41271431)

摘　　要：针对球形全景的匹配问题,提出了基于本质矩阵鲁棒估计的匹配方法。首先构建了球形全景的投影模型,通过推导球形全景的对极几何关系,得到本质矩阵的估计模型和对极线求解算法,再利用SIFT（scale invariant feature transform）算法进行特征提取和匹配,最后采用RANSAC（random sample consensus）算法,以改进的Sampson距离算子设置阈值,对本质矩阵进行鲁棒估计并剔除误匹配。实验结果表明,该方法切实可行,可提高球形全景匹配的精度。Aiming at matching problem of spherical panorama,this paper puts forward a matching method based on robust estimation of essential matrix.Firstly,we construct a projection model of spherical panorama,and get the estimation model of essential matrix and the solution algorithm of epipolar line through deducing the epipolar geometry relationship between spherical panoramas,then use scale invariant feature transform algorithm for features extraction and matching.Finally threshold value was set by using random sample consensus algorithm and the improved Sampson distance operator,robust estimation of the essential matrix was excluded,and the mismatches was removed.Experimental results show that this method is feasible and can improve the accuracy of spherical panoramic matching.

关 键 词：球形全景 本质矩阵 鲁棒估计 对极几何 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]