The Study Of Three-Dimensional Reconstruction Based On Large-Scale Disordered Images

With the improvement of abilities in computer image processing and the development of technologies based on images such as 3D printing,it is widely admitted that 3D reconstruction based on pictures and images has attracted a dramatic increasing interests among researchers in Computer Vision.Relavant technologies have also been applied universally in the fields such as Geographic Information Systems,Reverse Engineering,Virtual Reality,etc.Recently,many relevant studies are focusing on the issue about how to upgrade the authenticity of the reconstructed models while the other aspect is to improve the time-effectiveness of the reconstructed algorithm.This paper is attempting to present an advanced algorithm in 3D reconstruction which can reduce the reconstruction time and enhance the authenticity simultaneously referring to a large-scale disordered images.Finally,the results will show a prototype system based on large-scale disordered images for 3D reconstruction by designing and realizing.The main purposes of this paper include as following:(1)Present an image-fingerprint algorithm based on 2D gamma equilibrium to adjust the image brightness by 2D gamma value,then calculate the perceptual hash function of the adjusted image which can represent the image texture.The algorithm will be applied in data initialization and feature matching to filter out irrelevant images and improve the time-consuming of feature point matching rate at the same time,finally this paper is trying to obliterate the error point caused by irrelevant image.(2)Present an useless point deleting algorithm based on grid segmentation combined with octree algorithm to divide the data of point clouds,compare the point clouds between adjacent grids,use useless color's features and density features to remove useless points,and finally improve the point cloud quality.(3)Improve a 3D surface reshaping algorithm based on Delaunay triangulation,focus on solving the problem of 3D reconstruction such as sampling unevenness and large scale.Improved algorithm by introducing Voronoi polyhedron,the improved algorithm can effectively generate a convex shell of the 3D reconstruction point cloud.Experiments have shown the algorithm is reliable and effective.Compared with the point cloud model,the3 D surface model has a greater improvement in authenticity and in wider scenario.(4)Design and apply a 3D reconstruction system for a large number of disordered images then reconstruct the model of Terracotta Warriors by its graphic data.Experiments have shown that the system took a good reconstruction effect in large-scale disordered image's reconstruction with better time-effectiveness.