Study On Image-Based Real Time Walkthrough

Real time walkthrough in realistic complex virtual environments gradually becomes a hot topic in computer graphics, and it is increasingly in demands in various applications. One of its primary challenges is to find a way by which an environment could be well represented. With the increasing complexity of environment, traditional computer graphics based on polygons becomes hard to meet the requirement of real time rendering of a complex scene. On the other hand, two new techniques developed in recent years, image-based rendering and point-based rendering, come up to adopt new type of rendering primitives, and as a consequence, good results have been achieved in various applications. This thesis aims at embedding the two new techniques for real time walkthrough in a complex virtual scene.This thesis proposes an algorithm to build efficient representation for complex virtual environment from large set of reference images. Starting from multiple depth reference images, a hybrid representation of the complex scene will be automatically built up without knowledge of the original 3D models, and through this representation, a real time walkthrough in a large complex scene with no restriction to user's motion can be achieved.The main contributions of this thesis are: This thesis proposes an algorithm to build an efficient representation for large complex virtual environment from large set of reference images via pixel classification and OBB technique; This thesis proposes a hybrid representation of a complex scene by a combination of points and polygons. In more precise words in the approach, thelarge flat surfaces in the scene will be represented by textured polygons and the objects with high surface detail will be represented by points. The representation takes full advantages of traditional and cutting edge computer graphics techniques to achieve an efficient and packed representation for virtual environment; This thesis proposes a sampling rate comparison algorithm based on Jacobi matrix to eliminate redundant pixels among the reference images; This thesis proposes a texture reconstruction algorithm to reconstruct the corresponding textures of surfaces suitable for polygon-based representation. At the same time, holes which may appear in walkthrough will be analyzed and filled in the preprocessing stage; This thesis proposes a local reconstruction and resampling algorithm to generate a set of new samples which distribute more uniformly on surfaces and then they are organized by OBB-tree and normal vector clusters. Based on the efficient organization, several methods can be employed to improve rendering efficiency in walkthrough, such as cluster-based calculating of splat size, selecting proper detail level to render and so on; Based on the hybrid representation, this thesis proposes a real time walkthrough system for a large complex environment with no restriction to user's motion.