Research On Image-base Object Modeling And Rendering In Virtual Environment

The realism and efficiency of object rendering is one of the key problems for virtual environment. In complex virtual environments, geometry-based objects representation is always limited to the rendering complexity. This dissertation imports image-based modeling and rendering technology into object representation, aiming to improve the construction and rendering efficiency, and to enhance the virtual environment realism.In this dissertation, we first study the key technologies of view synthesis, and propose two object view-region models based on bi-view synthesis and tri-view synthesis. Then, we set up the image-based object representation method in virtual environment, and give some problems detail analysis and their solutions, which are associated with object realism. The original contributions of this dissertation include the following:(1) By study of view synthesis technologies, this dissertation proposes some corresponding key arithmetic. We propose a dispersed feature point detection arithmetic based on image segmention and connected subset reduction, the feature points are matched under multiple constraints in every bi-views or tri-views for tensor estimation. In image rectification process, we give analysis of rectification resampling and set up an evaluation function. To reduce the resampling effects, we propose the definition of rectification invariant and the corresponding correcting method. To improve the quality of dense match and view synthesis, a scanline segmenting and matching method is developed.(2) Two object view-region models based on bi-view synthesis and tri-view synthesis are set up. In the object view-region model, view space encircling the object is divided into several areas and sampled. Based on the research of view synthesis, we set up an object view-region model based on bi-view synthesis and tri-view synthesis, which constructs a view convex closure by triangles of every three adjacent cameral centers distributed in two sphere strip or a view convex polygon by line segments of every two adjacent cameral centers in same sphere strip. The object view-region model extents the viewpoint observing the object from dispersed point to continuous convex closure or polygon.(3) By analyzing technologies for image-based objects representation in virtual environment, the dissertation proposes some important problems contributed to realism and gives the solutions. In virtual environment, image-based object representation depends on the terrain, object position and viewpoint. In this dissertation, image-based object is represented by importing imposter for location, synthesizing correct view by view-region model for observation, and texture mapping of imposter for visual representation of image-based object.The most important three problems contributed to realism realism we consider are terrain matching, shadow generation and bounding volumes definition (for collision detection etc.). This dissertation gives those solutions for image-based objects, especially, we discuss the different terrain matching methods and their appropriate domain, propose a shadow generating arithmetic based on curved surface texture mapping in random lights, and give a try to define bounding volumes by object samples and imposter.(4) A prototype system is developed; the key technologies are verified and applied in projects. The primary arithmetic for view synthesis, object embedding and object realisticrepresentation are coded and the results are given in table or figure. A prototype system of image-based object modeling and rendering is set up and the core functions are packaged as view synthesis engine and object rendering engine, such engines are applied to correlative projects.Summarily, this dissertation proposes the fundamental theory, model and method of image-based object, and implements the image-based object rendering in virtual environment. Such research systematizes and enriches the image-based object theory; and gives some solutions for stereo vision and 3D reconstruction.