High-Quality 3D Reconstruction from Low-Cost RGB-D Sensors

Robert Maier

Abstract: This thesis explores the reconstruction of high-quality 3D models of real-world scenes from low-cost RGB-D sensors. Firstly, we propose a method for obtaining sharp model textures that uses super-resolution keyframe fusion for texture mapping. Secondly, we present a joint surface geometry and appearance optimization approach to recover fine-scale geometry and consistent surface colors. Finally, we introduce a method for correcting large-scale 3D reconstructions on-the-fly on pose graph updates.