Convex Variational Methods for Single-View and Space-Time Multi-View Reconstruction

und durch die Fakultät für Informatik am 07.02.2015 angenommen. This thesis investigates two special cases of 3D reconstruction: The reconstruction from only a single input image and the reconstruction over time from multiple-view image sequences. For both cases we propose several mathematical models which share the same basic idea: We compute a surface with minimal area that best fits the input data and suitable priors as the solution of a variational, convex optimization problem. Compared to state-of-the-art methods the proposed single-view reconstruction approaches require significantly less user input and yield competitive reconstruction results. For the space-time multi-view case, we show that the quality of a temporally coherent reconstruction improves substantially compared to time-independent approaches. A novel data fidelity term, the estimation and integration of surface normals and the integration of a novel, generalized form of connectivity constraints leads to reconstruction results that outperform the state of the art in both reconstruction accuracy and computation time