High-Quality Mesh Generation from 3D Scans for Surface Analysis

3D scanning has grown to become an important tool in a wide variety of applications. Still, acquiring high-quality 3D models using scanning technology is a challenging task. In this thesis, we present various ways that reduce the hurdles of existing 3D scanning pipelines with the ultimate goal of bringing this technology closer to the end user. To achieve this goal, we focus on three sub problems of traditional scanning pipelines: First, we present a novel algorithm that can be used to consistently orient the normals of huge point clouds. We achieve this by formulating the orientation problem as a graph-based energy minimization problem and applying out-of core methods. Second, we introduce interactivity into the scanning pipeline by presenting an online surface reconstruction method that produces high-quality semi-regular meshes. The resulting interactive pipeline is highly efficient because it reduces the turn-around between presentation of the final result and possible corrections by the user. Third, we develop a robust method to texture-map semi-regular meshes. This approach is based on a generalization of motorcycle graphs, which partitions arbitrary meshes into quadrilateral patches. These patches then serve as the parametrization domains of the texture atlas. Finally, we present an application from the area of cave science. The application is targeted at quantitatively and objectively assessing a cave's size. To achieve this goal, we present methods to analyze the structure of caves, especially to distinguish chambers from passages