基于深度学习的图像特征匹配方法  

Image feature matching method based on deep learning

作　　者：徐梦莹 刘文波[1,2] 郑祥爱 蔡超 XU Mengying;LIU Wenbo;ZHENG Xiangai;CAI Chao(College of Automation Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China;Key Laboratory of Non-Destructive Testing and Monitoring Technology for High-Speed Transport Facilities,Ministry of Industry and Information Technology,Nanjing 211106,China)

机构地区：[1]南京航空航天大学自动化学院,江苏南京211106 [2]高速载运设施的无损检测监控技术工业和信息化部重点实验室,江苏南京211106

出　　处：《传感器与微系统》2022年第7期61-64,共4页Transducer and Microsystem Technologies

基　　金：国家重点研发计划资助项目(2018YFB2003304);国家自然科学基金资助项目(61871218);南京航空航天大学研究生创新基地(实验室)开放基金资助项目(KFJJ20190306)。

摘　　要：针对同一场景或同一物体的两组或多组图像的匹配问题,提出了一种基于深度学习的图像特征匹配方法。首先用SuperPoint网络框架提取图像特征点,在最近邻次近邻比值法的基础上通过网络运动统计(GMS)算法区分正确匹配点和错误匹配点,最后采用随机抽样一致性(RANSAC)算法进一步剔除误匹配点对。实验结果表明:所提算法在图像发生光照以及视角变化时平均匹配确正确率达到95%以上,具有较好的匹配识别率和鲁棒性。For the matching of two or more groups of images of the same scene or the same object,an image feature matching method based on deep learning is proposed.Firstly,the image feature points are extracted by using the SuperPoint network framework.On the basis of the ratio of the closest neighbor and second closest neighbor method,the correct matching points and wrong matching points are distinguished by grid-based motion statistics(GMS)algorithm.Finally,RANSAC algorithm is used to further eliminate the mismatched point pairs.The experimental results show that the average matching accuracy of the proposed algorithm is over 95%with image changes in illumination and blurring,and the proposed algorithm has good robustness and stability.

关 键 词：深度学习 SuperPoint 网格运动统计算法 随机抽样一致性算法 

分 类 号：TP391.5[自动化与计算机技术—计算机应用技术]