Research Of Oblique Image Matching Based On Local Affine Invariant Features

With the development of "smart cities",the social demand for 3D modeling is increasing.oblique photogrammetry is one of the main methods of 3D modeling.However,due to the large affine inclination angle and severe occlusion in the oblique image,the original image matching algorithm has problems such as a small number of matching points,low matching efficiency and poor reproducibility in oblique image matching.Local affine invariant features have better adaptability to image translation,rotation,illumination changes and viewing angle changes,and local features can better handle occlusion problems than global features,so this paper proposes a fusion Hessian-Affine feature and MSER feature oblique image matching algorithm.First extract Hessian-Affine features and MSER features,refine and optimize MSER features,use SIFT descriptors to generate feature vectors,and obtain coarse matching results under two-way nearest neighbor constraint and the nearest/next distance ratios constraint,and finally remove the gross errors by RANSAC algorithm to complete fine matching.Through experimental analysis,the number of feature points obtained by the algorithm in this paper is slightly lower than the ASIFT algorithm,but the algorithm takes only 1/10 of the ASIFT algorithm,and the matching accuracy is higher than the ASIFT algorithm and the SIFT algorithm.In addition,compared with the single-feature image matching algorithm,the uniformity and richness of correctly matched point pairs have been greatly improved.The uniform distribution of feature points helps to improve the accuracy of the transformation model between the images to be matched,thereby providing stable and uniform point pairs of the same name for subsequent dense matching.The specific work can be divided into the following aspects:(1)Introduce the research status of image matching,and introduce the basic principles and processes of SIFT algorithm,MSER algorithm,Hessian-Affine algorithm and ORB algorithm in detail;(2)Use the graffiti data set for the oblique image matching experiment,and analyze experimental results of RANSAC algorithm,VFC algorithm and GMS algorithm for eliminating gross errors.The experimental results show that the RANSAC algorithm is generally better than VFC algorithm and GMS algorithm for eliminating gross errors,RANSAC algorithm has relatively high accuracy and shortest average time.(3)A oblique image matching algorithm combining Hessian-Affine features and MSER features is proposed.The non-maximum suppression algorithm is used to refine and optimize the MSR feature regions,and the MSER features are normalized to generate feature vectors using SIFT descriptors.And verify the algorithm of this paper on the graffiti data set and the real oblique image.