Surface Reconstruction from Noisy Point Clouds

Boris Mederos

Keywords:

In this work we study the problem of surface reconstruction from noisy data sets. We present some original methods to face this problem: We present a new algorithm for smoothing point clouds in the presence of noise, able to eliminate noise, while keeping the salient features of the original model. In order to structure the resulting point clouds we introduce a new advancing front surface reconstruction algorithm, which handle surfaces with boundary, non-uniform sampling and arbitrary topological types. Also a new refinement method that refines the initial triangulation, producing a refined mesh adapted to the geometry of the unknown surface. Finally we study the {\em power crust} algorithm of Amenta et al. \cite{AMENTA4}, presenting a variations of the {\em power crust} adapted to noisy samples and we prove that the reconstructed surface is isotopic to the unknown surface.

Anexos:

• thesis_boris.ps.gz