A study of Symmetric and Repetitive Structures in Image-Based Modeling

Abstract

Creating photo-realistic 3D digital models from street-view imagery has many important applications and involves fundamental vision problems. We investigated the paradox of having similar or repetitive structure in the input image data. In general, prior knowledge of structure regularity helps with the efficiency and quality of image-based-modeling; however, spurious camera geometries due to appearance ambiguity arising from similar structure can lead to algorithm failure in structure-from-motion, especially for unordered image collections. In this dissertation, we made a detailed survey on 3D reconstruction methodologies and proposed a novel objective function based on `missing correspondences? to evaluate the optimality of a 3D reconstruction. An efficient algorithm is designed for optimization. We also investigated the problem on automatic detection of repetitive structures in the recovered scene and proposed a method to jointly analyze images and 3D point clouds to symmetric lattices. Finally, symmetry is further exploited for a novel camera calibration method and an interactive 3D modeling system working with a single input image.