Key Technologies Research Of City Three-dimensional Scene Reconstruction Based On Oblique Images

Oblique image-based 3D scene reconstruction of city is a cross research hotspot in photogrammetry and remote sensing, computer vision, graphics learning etc., and can be used for urban construction and management, public life and travel etc. It is a complicated and comprehensive charming techniques and sciences. This paper selected a number of key techniques in three typical problems,3D model construction technique based on oblique images, seamless texture mapping technique, and triangle texture storage, for in-depth study. In this paper, the specific research contents and results are as follows:(1) 3D reconstruction based on oblique images. This paper completed a set of 3D surface mesh model reconstruction strategy based on oblique images. At first, the paper put forward a corner feature extraction method with affine invariant, and verify the effectiveness of the algorithm by comparing with the famous algorithms of Harris and Harris-Affine, and put the corner points into the seed point set of dense matching to guide and constraint the matching process. Then, according to the characteristics and insufficiency of original PMVS algorithm, the paper proposed a improvement PMVS dense point cloud stereo matching algorithm, it under a adaptive resolution point cloud reconstruction strategy, through real-time optimization and adjustment of dense matching, to improve the precision and efficiency of the dense point cloud reconstruction. At last, the paper adopted the Poisson triangle reconstruction method to generate the surface mesh model, at the same time, to simplify the original triangle mesh model by re-calibrating the dense point cloud.(2) Radiation correction for texture images. This paper completed the design and experimental analysis on fogging algorithm based on dark channel and shadow detection and elimination algorithm based on color space transform. In the fog algorithm based on dark channel, the paper calculated the transmittance map on different resolution, accelerating the calculation speed; simultaneously, introduction a automatic halftone algorithm to processing of gradations of color image, to improve the images visual effectives under different lights and angles. In addition, the function of the halo and white balance coefficient were considered in the camera response function optimization estimation, and the error of camera system was corrected by it. Finally, the methods reduce the phenomenon of color jumps and figures in the texture mapping.(3) Optimization and storage for triangle textures. The paper completed a design and experiment on the best triangle texture selection and clustering strategy. First of all, a visible triangle textures list be established by screening principle, effective shade detection and adaptive subdivision of part occlusion triangular. Then, the paper under the MRF with Graph cuts optimization algorithm to solve the problem of selecting the best texture for every triangle on the mesh model, and generating the seamless texture mapping model. Moreover, in order to save the memory consumption in the process of computer texture rendering, an optimization storage strategy of texture atlas be used. This paper proposed a fast and available method to boost the space utilization of texture atlas, and the method was proved to be effectively.