摘要
针对传统ORB(Oriented Fast and Rotated BRIEF)算法误匹配率高、抗视角变换能力弱等问题,提出一种融合交叉验证ORB和一致空间映射的快速图像匹配算法。在ORB算法的框架下分别检测模板图像和目标图像的特征点集,并利用每个特征点的描述符建立两个特征点集之间的对应关系;利用每对特征点邻域的支持点集交叉验证对应特征点对的准确性;在粗剔除低准确性的对应特征点对后,通过一致空间映射对经过筛选的特征点集进行稳健的非线性匹配。实验结果表明,该算法优于当前主流的图像特征匹配算法,并且在处理大视角变换和现实场景下的图像特征匹配问题时稳健性较强。
Aiming at the problems of high mismatch rate and weak anti-viewing ability of traditional ORB algorithm,we propose a fast image matching algorithm fusing cross-validation ORB and consistent spatial mapping.It detected the feature point set of the template image and the target image in the framework of the ORB algorithm,and established the corresponding relationship between the two feature point sets by using the descriptor of each feature point.Afterwards,we used the support point set for each pair of feature point neighborhoods to cross-verify the accuracy of the corresponding feature point pairs.Then,after roughly eliminating the corresponding feature point pairs with low accuracy,robust non-linear matching of the selected feature points set was carried out by consistent spatial mapping.Experiments show that the proposed algorithm outperforms the current mainstream image feature matching algorithm.And it has strong robustness in dealing with the problem of image feature matching under large viewing angle transformation and real scene.
作者
周光宇
钟汉生
章晓敏
Zhou Guangyu;Zhong Hansheng;Zhang Xiaomin(College of Digital Technology and Engineering,Ningbo University of Finance and Economics,Ningbo 315175,Zhejiang,China)
出处
《计算机应用与软件》
北大核心
2020年第9期176-182,共7页
Computer Applications and Software
基金
浙江省自然科学基金项目(LQ18F010008)。
关键词
计算机视觉
匹配点对
交叉验证ORB
一致空间映射
非线性匹配
Computer vision
Matching point pair
Cross-validation ORB
Consistent spatial mapping
Nonlinear matching