Line Matching Based On Point Correlation Description Across Views

As one of the most basic problems in image feature matching,line matching has a verybasic and important position in the field of computer vision,which plays an important role in3 D reconstruction,pose estimation and camera calibration.Using the line-points relationship as descriptor to perform line matching is less affected by the structural and texture information,but the accuracy of point matching has an important influence on the line matching results.In order to improve the accuracy of line matching,this paper introduces a more accurate method namely Descriptor-Nets(D-Nets)for feature point matching,and then the results of feature point matching are taken to lead the work of line matching.We use the matching feature points in the line neighborhoods and the line to construct the line-points projective invariant,and the similarity between the line neighborhoods can be computed through the invariant,and finally obtain the matching lines.Experiments show that the higher accuracy of D-Nets method makes line matching method based on line-points relationship have better matching results.In the multi-viewpoint line matching process,the invariant is an effective mean of describing that the line remains unchanged under the projective transformation.Using the intersection points will make the results of point matching more stable.Therefore,this paper uses D-Nets to match the intersection points of lines,and then the results of intersection point matching are taken to lead the work of line matching.Since lines are usually located on the plane or the edge,we use the matching intersection points to obtain the homography matrix between the images,which can be used to get more potential matching lines.Compared with the state-of-the-art method,our method is robust to various image changes,and can get better performance.