Research On High Quality Texture Mapping Method In 3D City Scene

With the development of Internet technology and the popularization of intelligent equipment,the digital age based on geographic information data is quietly changing the way people live and learn.In this context,the construction of a three-dimensional visualization model has important practical significance for enhancing people's cognition of geoscience information and building a bridge for multidisciplinary communication and cooperation.The3 D reconstruction technology based on oblique photography is an important method to generate geographic information data.Among them,texture mapping is an important part of3 D reconstruction.The authenticity and efficiency of 3D products are limited due to illumination,exposure changes,texture discontinuity,data redundancy,and image scaling.Therefore,texture mapping of the model is A very challenging subject.The main research contents of this paper are as follows:1)Since oblique images have the characteristics of large viewing angles,inconsistent proportions,uneven exposure changes,and large amounts of data,how to select high-quality textures from multi-view images to map a 3D model is a major difficulty.This paper improves on the texture selection strategy based on the Markov model proposed by Waechter M et al.The basic idea is: the data items of the Markov model comprehensively consider the average color gradient of the image,and the area size of the triangle surface projected on the image.In order to reduce the inherent texture distortion in the process of image mapping to the model surface,the smoothing item is constrained by considering the texture continuity,and finally the graph cut method is used to solve the problem.High-quality texture sources are selected for each triangle on the 3D model.2)Aiming at the occlusion problem in large urban scenes,this article makes two improvements.One is to adjust the realization of the deep detection algorithm and accelerate it in combination with the characteristics of GPU;the second is to propose an occlusion judgment method based on a deep learning model,First use the deep learning model to detect moving obstructions(vehicles,pedestrians,etc.)in the de-distorted image to generate a mask,and then set the weight of the obscured part to 0 when scoring the visual image of the triangle to avoid Select the texture image with occlusion,and solve the distortion problem caused by the wrong texture selection from the source.3)Aiming at the problems of chromatic aberration,obvious gaps between textures and inefficient processing after 3D reconstruction of oblique images,this paper proposes a fast and efficient method of seamless texture homogenization.First,perform image enhancement preprocessing on the input texture image,and adjust the overall color of the texture by establishing constraint equations for all vertices and edges of the model in the homogenization stage of texture mapping;then,after texture expansion,a global texture block with no color difference is obtained.;Then the texture block is edited with reduced resolution Poisson;finally,the texture block is arranged to obtain a compact texture image.The texture on the 3D model has no color difference between blocks,the texture transition at the seam line is natural,and the calculation speed is faster.At the same time,it enhances the readability of the 3D model and improves the texture display effect of the 3D model.