Research On Automatic Texture Mapping With Multi-View Images And Scene Visualization In Complex Urban Environment

A real three-dimensional city model has both high-precision geometric shape and surface texture property.As textures increase the details of object surface,they highly enhance the reality and comprehension of visual 3D scenes[1].With measuring equipment accuracy and efficiency,and post-data processing technology maturity,there are more practical applications in urban building construction and restoration of the precise three-dimensional geometric shapes,but less researches and breakthroughs in texture reconstruction of the large-scale three-dimensional model.Based on the brief introduction of 3D reconstruction process of digital city model,this paper mainly focuses on how to accurately map the real images taken by digital cameras back to the target three-dimensional geometric model after restoring the three-dimensional geometric shapes of urban buildings and structures,and how to organize and render the 3D scene with large data volume in limited computer resources.At last,the paper puts forward a technical scheme with multi-view image and automatic texture mapping,and an organization and visualization method in three-dimensional scene with large data volume Program.Specific research content as follows:(1)This paper systematically introduces the processes of automatic 3D reconstruction using multi-view images and briefly summarizes current researches of key technologies applied in the phase of 3D camera reconstruction,image orientation,dense matching,surface reconstruction,texture mapping and et al.The paper also presents the relationship between the results and the automatic texture mapping,and puts forward relevant ideas and processes of using the multi-view image to carry out the automatic texture mapping of the urban three-dimensional geometric model.(2)This paper analyzes two kinds of multi-view image optimal texture distribution techniques based on multi-factor weighting and Markov random field optimization.On the basis,the initial texture distribution of the multi-image automatic texture mapping scheme has been selected.Under the initial texture distribution scheme,the paper concretely explains the problem of texture illumination and color inconsistency at the texture joint,and proposes a method that is used to solve the global color by solving the minimum value of the quadratic function to eliminate the inconsistency of multi-image color and light which are caused by different shooting angles and conditions.(3)This paper applies the Poisson Image Editing algorithm to study how to further eliminate texture seams.Global Texture Color Discontinuity Compensation cannot completely eliminate all texture seams caused by inconsistencies in texture color.Visible seams still exist in adjacent texture blocks that use different images as textures.Therefore,the adjusted results of global color consistency are further processed locally.(4)This paper explores how to pack a small texture block into a large texture map,and proposes a greedy nesting algorithm for small texture blocks.(5)This paper presents a scene organization and visualization scheme for large-scale real three-dimensional city.As the area of city scene is usually as large as tens of square kilometers,the amount of geometric data and texture data are extremely large for a real three-dimensional city,it is important to presents a scene organization and visualization scheme to meet the demand of browsing the models in a ordinary computer.