Signed Distance Fields for Rigid and Deformable 3D Reconstruction

This thesis addresses the tasks of 3D reconstruction of rigid and deformable objects and scenes, scanned with a single handheld RGB-D sensor. To this end, we propose an implicit-to-implicit alignment scheme between pairs of signed distance fields. We first apply it for reconstructing household objects, achieving higher trajectory and model precision than existing approaches, and then adapt it to larger scenes. Finally, we extend it to reconstruct moving objects that undergo topological changes and large motion, which are challenging for previous methods.