A High-realistic Texture Mapping Algorithm Based On Image Sequence

Image based 3D reconstruction method has been widely applied in many fields because of its fast reconstruction speed,high accuracy and operability.The 3D reconstruction process based on image includes,image acquisition,feature points detection and feature point matching between images,using camera calibration principle to restore the scene,CMVS/PMVS algorithm be used to dense reconstruction 3D point cloud model with noise.Thus the point cloud model will not only contain a lot of noise,and there is a big distance between the data points,which makes the model with low precision and fidelity.In view of the reconstructed point cloud model,the traditional methods need to go through many steps,such as denoising,gridding,repairing flow pattern edges,parameterization,projection texture,and so on,so that we can get a high fidelity 3D model.In this paper,the texture mapping method based on the point cloud model which be reconstructed by images is studied.The main research content and work are as follows:1.We discussed the related technology and characteristics of image based 3D reconstruction algorithm,and used an improved bidirectional KNN matching algorithm to make the matching results more precise in the process of feature point matching.The status of the research on texture mapping at home and abroad is illustrated,and the advantages and disadvantages and applicability of the existing algorithms are analyzed.2.For large-scale complex scenes,the input image sequence is usually very long.This paper explores the spatio-temporal adaptive sampling method to select the appropriate key image for subsequent texture mapping optimization,so as to reduce the processing time and improve the quality of the generated texture map.First,the input image sequence is calculated in the time domain,and then the image with a relatively low degree of ambiguity is selected for sampling.Finally,the uniqueness of the previously obtained key image with the smallest ambiguity is calculated.After that,the redundant information can be further reduced and the key images withsmaller differences can be removed,thus the processing of subsequent texture data can be reduced.3.Proposed texture mapping algorithm for realistic 3D model reconstruction of a use of color and texture,the algorithm is mainly divided into three steps: the first step is to use the Bundler to obtain the correspondence between the three-dimensional points of the space and the two dimensional points on each image;the second step is to select the most appropriate mapping texture blocks for each triangle.In order to avoid the redundancy of input pixel data,we texture selection before texture mapping,only preserving texture corresponding regions,so as to speed up texture mapping speed.The third step is to optimize the texture consistency for the mapping results,so as to avoid the color difference at the boundary of the texture.