| 3D reconstruction mainly uses two or more images to restore 3D information of images and reconstruct 3D scenes from different perspectives.It has been widely used in the fields of digital protection of Ancient Architectural Images and automatic driving.Feature detection and matching,as one of the key steps of 3D reconstruction,directly affect the result of 3D reconstruction.However,for the image with complex structure and repeated texture,the feature detection and matching process has the problems of low accuracy and poor robustness.Aiming at the above problems,combined with the geometric structure between the image feature lines and the characteristics of strong expression ability,this paper conducted an in-depth study on the detection and matching method of the image feature lines.The main research results are as follows:1.To reduce the breaklines and high false detection rate in the line detection of complex scenes and repeated objects with textures,a feature line detection algorithm named BPC＿GF is proposed,which combined the idea of edge detection.Firstly,the improved Canny edge detection algorithm is used to detect the attributes of image edge points.Secondly,the aim point is determined from the edge points,the concept of basic blocks is introduced,and different types of basic blocks are generated by the greedy algorithm.Then the basic blocks of the same type are grouped according to the angle deviation of the main direction and the spatial distance constraints,and fusion the the basic blocks generate candidate feature lines.Finally,the improved Helmholtz principle is used to eliminate the false characteristic lines and the accurate characteristic line set is obtained.2.To solve the problem of high mismatching rate under the influence of multiple factors such as scale transformation,perspective change,noise and illumination,a feature line matching method based on multiple constraints(LMC＿LSD)was proposed by combining the scale invariant descriptor SIFT and feature line detection algorithm LSD.Firstly,the supporting region is determined for each characteristic line detected.Secondly,in the support region of the feature line,the feature points on both sides of the feature line and the two endpoints of the line are matched with the same name.Then,the kernel constraint was applied to the feature line,and the supporting region of the matching line was determined according to the position of the point with the same name and the kernel line,and the similarity constraint of Angle,distance and gray level was used to get the correct matching of the feature line.3.Based on the BPC＿GF and LMC＿LSD methods,a prototype system of image feature line detection and matching is designed and implemented with the image of Ancient Architectural Images as the object.The prototype system mainly includes image preprocessing,Canny edge detection,SIFT feature detection,feature line detection,feature line matching and other functions.The system provides technical support for 3D reconstruction of Ancient Architectural Images. |
