Study On Several Issues Of Matching And Reconstruction For Oblique Multi-view Stereo Images

With the development of photogrammetry and computer technology,3D reconstruction technology has been widely applied in many fields such as digital city construction and restoration of cultural relics.Meanwhile,the rapid development of UAV makes the use of UAV aerial images for 3D reconstruction possible.However,there are various transformations such as rotation,affine and scale in the images of UAV,which increases the difficulty of image matching.Image matching is one of the critical steps in 3D reconstruction,in order to complete the 3D reconstruction of the object,it is of vital importance to study the image matching of UAV.Due to the existence of “blind area” when the UAV acquires the images,the generated point cloud appears “hole area”,which will affect the accuracy and structural integrity of the subsequent reconstruction model directly,therefore,it is of great significance to study how to repair the “hole area” problem of point cloud.Aiming at this practical problem,this paper researches on several key technologies in the process of 3D reconstruction.The details are as follows:(1)In order to improve the reliability of initial sparse matching and the number of matching points,a feature matching method based on Scale Invariant Feature Transform(SIFT)and DAISY algorithm is adopted.Firstly,the method uses the SIFT algorithm to extract feature points and then based on the DAISY algorithm to describe the feature,and based on the nearest neighbor ratio method to match,finally,the method uses the Random Sample Consensus(RANSAC)algorithm to eliminate the mismatch.Compared with using SIFT algorithm alone,this method improves single using description method of SIFT feature points in Structure from Motion(SFM)algorithm,it turns out that the number of correct matching points obtained by this method is more and the correct matching rate is higher.(2)In order to improve the structural integrity of 3D reconstruction models,a twostep successive three-dimensional reconstruction strategy is constructed based on camera calibration.Firstly,this strategy implements sparse reconstruction based on reconstructed SFM algorithm to generate three-dimensional sparse point cloud.Then the Clustering Views for Multi-view Stereo(CMVS)and Patch-based Multi-view Stereo(PMVS)algorithm based on region growing is applied to further dense reconstruction built on sparse point cloud.The experimental results show that the strategy can get denser and more complete three-dimensional point clouds.(3)Aiming at the problem of point cloud “hole” caused by “Blind area” of photography in low-altitude image acquisition,a hole repairing scheme based on the ground close-range image is adopted.Acquiring the sequence images of the hole area via the handheld camera,the dense reconstruction of point cloud is obtained by the aforementioned method,then the registration and fusion of two point sets are realized based on the Iterative Close Point(ICP)algorithm,thus filling the hole area,obtaining more complete reconstruction of target point cloud.Finally,the dense point cloud is processed by triangular meshing and texture mapping,a complete 3D model is obtained.