Surface Reconstruction from Imprecise 3D Points

Kurzfassung

In applying optical methods for automated 3D indoor modelling, the 3D reconstruction of objects and surfaces is very sensitive to both lighting conditions and the observed surface properties which ultimately compromises the utility of the acquired 3D point clouds. This works presents a statistical method for surface reconstruction applying robust statistics from the compressive sensing discipline. It is shown that 3D point clouds with outliers, strong noise and varying sampling density can be processed by the presented method without manual interaction. The resulting surface models are stored in implicit functional form and may be processed by common rendering algorithms such as marching cubes to deliver polygonal models of arbitrary resolution.