一种渐进式的鲁棒特征匹配方法  

A Progressive Method for Robust Feature Matching

作　　者：马伟乐 肖腾[1] 杨军星 梅熙[2] 王义[2] 邓非[1] MA Weile;XIAO Teng;YANG Junxing;MEI Xi;WANG Yi;DENG Fei(School of Geodesy and Geomatics,Wuhan University,Wuhan 430072,China;China Railway Eryuan Engineering Group Co.Ltd.,Chengdu 610031,China)

机构地区：[1]武汉大学测绘学院,湖北武汉430072 [2]中铁二院工程集团有限责任公司,四川成都610031

出　　处：《测绘地理信息》2023年第6期56-61,共6页Journal of Geomatics

基　　金：四川省科技计划(2019YFG0460)。

摘　　要：针对影像特征匹配初始结果中存在大量粗差的问题,提出了一种渐进式的鲁棒特征匹配方法,分为种子特征匹配点提取与匹配点扩展两步。其核心思路是先利用特征点的局部几何一致性进行误匹配剔除,提取可靠的种子匹配点;然后,利用由种子匹配点估计出的全局几何模型,对非种子匹配点进行验证,完成匹配点的扩展,得到最终的鲁棒特征匹配结果。实验采用多种数据集对本文方法进行定性、定量分析,并与其他鲁棒特征匹配方法进行比较分析。结果表明,本文方法可以有效地保证特征匹配的精度,同时提高匹配点的数量。To deal with a majority of outliers in raw results of feature matching,a progressive method for robust feature matching is proposed,which is divided into two steps:seed matches extraction and then expansion by geometric verifica⁃tion.This method first uses the constraints of local geometric consistency between points and their neighbours to extract reli⁃able seed matches.Then obtain the global geometric model through these seed matches and verifies the remaining match⁃es,which leads to the expansion of matches.To qualitatively and quantitatively evaluate our method,various datasets are tested and some state-of-the-art robust methods are com⁃pared.The results show that our method can effectively en⁃sure the accuracy of feature matching and increase the number of matches.

关 键 词：渐进式策略 鲁棒特征匹配 局部几何一致性 

分 类 号：P237[天文地球—摄影测量与遥感]