直线对几何特征约束的近景影像特征匹配

Line Matching for Close-range Images with Geometry Features of Line Pairs

针对直线匹配过程中线特征提取不完整以及缺乏考虑邻近直线相关性,提出一种直线对几何特征约束的近景影像特征匹配方法:首先,利用直线检测算法(line segment detector,LSD)对参考影像和搜索影像进行直线提取,并对提取的直线进行特征编组获得特征直线对;然后,将特征直线对作为匹配基元,依次运用双重核线约束、仿射变换约束、最优相似性原则确定同名直线对;最后,根据同名直线对中2条直线端点与其邻域内同名点的距离关系拆分同名直线对,利用仿射变换约束整合拆分过程中对应不一致的单直线,得到最终匹配的同名单直线。为了验证该算法的有效性与可靠性,选取4组人工建筑物特征变换近景影像进行实验,验证了该算法在直线匹配过程中具有一定的稳健性和普适性。

The line matching algorithm for close range images with geometry features of line pairs is proposed in this paper,to seek the solutions that incomplete extraction of lines feature and lack of relationships between matching lines and adjacent lines in the process of line matching.Firstly,lines are extracted from the reference images and searching images by using LSD(line segment detector)matching algorithm,at the same time,grouped line pairs are obtained by the basic relationships such as distance and angle between the lines.Then,the algorithm makes the grouped line pairs as matching primitives,and matches these grouped line pairs by using corresponding epipolar constraint,linear affine transformation constraint and optimum similarity principle in order.Finally,it resolves the corresponding line pairs into two pairs of corresponding individual lines according to the distance relationship between the endpoints of two lines and their close-by corresponding points,and obtains one to one matching results after fitting and checking.In order to verify the effectiveness and reliability of the algorithm,this paper adopts four groups of close-range images with typical features from the online public image database.Through the comparative analysis of the experimental results,the algorithm shows certain robustness and universality in the process of line matching.