Modeling with panoramic image network for image-based walkthroughs

Abstract

Virtual walkthrough is now increasingly applied in more and more applications, like computer games and virtual touring. Computer Graphics techniques can be used to create a virtual environment by defining the geometric and lighting properties. But the construction of such models requires intensive labor work. On the other hand, Computer Vision attempts to reconstruct 3D models from images with correspondence matching. These models have the potential to be recovered automatically and are photorealistic. However, dense and accurate correspondence matching is difficult for occluded and textureless areas. These limitations impede the photo-realistic image-based walkthroughs from becoming competent. This thesis presents a novel method for modeling an environment by recovering camera parameters and geometric proxy from a panoramic image network. We propose an automatic camera parameter estimation method for panoramic image network. Instead of relying on scene structures, only point correspondences are required. This allows our method to be applied to environments with different complexities. We have also developed a method to ensure the global consistency of the camera parameters among the panoramic images. Based on the estimated camera parameters, we then attempt to recover the geometric proxy. Two methods are proposed under different approaches. An area-based method is suggested first to sparsely extract feature points and then recover global 3D meshes and local 2D meshes. Another phase-based method is proposed to recover the geometric proxy in a multiresolution approach with wavelet decomposition. Region with different levels of texture in the same environment can be handled. We further compare these approaches in terms of texture handling, coherence and correctness of depth information as well as object continuity.

Date of Award	15 Jul 2005
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Chong Wah NGO (Supervisor) & Rynson W H LAU (Co-supervisor)