摘要
针对单直线几何属性的弱稳定性及“非一对一”匹配结果难以检核问题,提出一种结合线对几何特征及单线描述符约束的直线匹配算法。所提算法将邻域内满足一定几何条件约束的两直线构建线对,作为一个整体进行匹配。匹配过程中首先利用线对中两直线交点的核线约束匹配候选范围,然后逐一采用线对内角度、线段间距离比、线对邻域辐射信息三种属性特征对匹配候选进行筛选,最后根据三角形区域灰度相似性确定最终匹配对。检核过程中首先根据直线与核线夹角及直线斜率建立同名线对中单直线的对应关系,然后每组同名线对分裂得到两组同名直线,在此基础上对应地建立两直线梯度描述符并计算两直线描述符间的相似性,最后结合共线几何和描述符相似性对匹配结果进行检核,剔除错误匹配,并对结果中的共线直线进行合并,得到一对一的同名直线。选取典型纹理特征的航空影像和不同变换类型的近景影像进行参数分析及直线匹配实验,实验结果表明,所提算法匹配正确率较高,在纹理相似、视角变化、旋转变化、尺度变化、光照变化的复杂场景下匹配正确率均高于95%,具有较好的鲁棒性,且可有效解决因直线断裂等原因造成的复杂匹配关系难以检核问题。
Objectives:In the field of line matching and checking,the geometric attributes of the individual line are weakly stable,and the non one-to-one matching results are difficult to be checked.To address the problems above,a line matching algorithm based on pair-wise geometric features and individual line descriptor constraints was proposed.Methods:For matching process,first,two line segments which satisfied certain geometric constraints in the neighborhood were grouped into a line pair and matched as a whole.Second,the epipolar constraint of the intersection in the line pair was used to determine the matching range.And characteristic angles within line pairs,distance ratio between line segments and the radiation information of the neighborhood of line pairs were used to narrow the range of the matching candidates.Finally,the final matching pairs was obtained by calculating the gray similarity of triangle region.For checking process,initially according to the angle of line with epipolar and the slope of line,the corresponding relation between individual lines was established.Moreover,each corresponding line pair was split into two groups of corresponding individual lines.Then,descriptors were established for two single lines in each group and the similarity between the two line descriptors was calculated.Eventually,collinear geometry and descriptor similarity were combined to check the matching results,eliminate the false matches.Collinear lines in the results were merged and one-to-one matching results were obtained.Results:Aerial images with typical texture features and close-range images of different transformation types were selected for experiments.The results demonstrate that the proposed algorithm has a high matching accuracy.The matching accuracy rate is higher than 95% in complex scenes with similar texture,perspective change,rotation change,scale change and illumination change.Conclusions:The proposed algorithm has good robustness to different types of images,and also has advantages for complex line matching
作者
刘肃艳
王竞雪
沈昭宇
王强
LIU Suyan;WANG Jingxue;SHEN Zhaoyu;WANG Qiang(School of Geomatics,Liaoning Technical University,Fuxin 123000,China;Faculty of Geosciences and Environmental Engineering,Southwest Jiaotong University,Chengdu 611756,China;School of Geography and Environmental Sciences,Tianjin Normal University,Tianjin 300387,China)
出处
《武汉大学学报（信息科学版）》
EI
CAS
CSCD
北大核心
2023年第6期936-949,共14页
Geomatics and Information Science of Wuhan University
基金
国家自然科学基金(41871379,42071343,42001412)
辽宁省兴辽英才计划(XLYC2007026)
辽宁省应用基础研究计划(2022JH2/101300273)。
关键词
直线匹配
组直线匹配
结果检核
共线几何
直线描述符
line matching
pair-wise line matching
results checking
collinear geometry
line descriptor
