Research On Key Technology Of Multi-view Oblique Aerial Image Matching

Compared to conventional single-len photogrammetry,multi-view oblique aerial photogrammetry which can obtain the textures of buildings from different sides,provides abundant image texture data for image matching and 3D modeling.Image matching is the core stage of digital photogrammetry,whose quality has direct influence on the precision of aerial triangulation,DSM,3D modeling and etc.This paper is mainly focused on technologise and features of oblique photogrammetry,matching constraint,feature-based matching and dense matching.The main work and innovation are as follows:1.Aerial oblique image acquisition technology,oblique image characteristics,difficulty of image matching,multi-view matching method and etc are introduced.Aiming at the advantages of massive multi-view oblique image data,the advantages of multi-view oblique image matching are discussed.The rich image data provides a lot of texture information for matching,which can improve the accuracy of image matching.2.The basic principles of DoG,Harris,SIFT and ASIFT are discussed,the characteristics of these four algorithms are compared.The single constraint and the epipolar geometric constraint are used for the oblique image matching,and the precision of the single constraint and the epipolar constraint are analyzed and compared.In order to solve the shortcomings of time complexity of ASIFT algorithm,ASIFT parallel processing is designed,and the algorithm improves the matching efficiency.3.A multi-resolution hierarchical matching PMVS algorithm based on SIFTGPU is proposed.Firstly,the Harris and DoG algorithms in PMVS are replaced by SIFTGPU algorithm with affine transformation and parallel acceleration to ensure the stability and density of the initial seed points with the angle of view.Secondly,the initial seed point is the basis of the spread of dense cloud,and its advantages and disadvantages directly affect the precision of dense reconstruction.Therefore,In this paper,the SIFTGPU algorithm based on the on-line constraint and the single-constraint combination is used to improve the PMVS processing efficiency and the reconstructed point cloud density.Finally,according to the different resolution of the image provided by the PMVS algorithm,the images are adaptively matched with different resolutions,and the resolution is different according to the requirements.The multi-resolution hierarchical matching is realized,and the algorithm improves the number of dense matching points and calculate the efficiency.4.A semi-global match based on minimum spanning tree algorithm is proposed.Because of the different image texture information of different images,in this paper,the image similarity detection is carried out by using the combined matching cost of pixel gray difference square and pixel mutual information.And exploit the idea of minimum spanning tree,to match the cost of aggregation,to find the best matching cost,the parallax graph generation.Through the experimental analysis of the standard database and the aerial oblique image,it is proved that the algorithm has a good matching effect.