Speaker: Felix Neumann
Abstract
Representing geometric shapes implicitly through signed distance functions has advantages in many aspects of geometric processing. Furthermore, there exist sophisticated techniques for generating signed distance functions based on scans of real-world objects. However, in many use cases explicit surface representations like triangle meshes are required. The preservation of sharp features like edges and corners is a key concern when generating explicit surface descriptions from implicitly defined volumetric data.
Most feature extraction methods for signed distance functions do not consider uncertainty. In practice, however, scans of real-world objects contain uncertainties due to sensor noise or view occlusions. This Master’s thesis considers signed distance functions whose values are modeled as Gaussian-distributed random variables as input data and aims to extract descriptions of sharp features present in the underlying shape.
