Construction of Implicit Surfaces from Point Clouds Using a Feature-based Approach

Abstract

We present a novel feature-based approach to surface generation from point clouds in three-dimensional space obtained by terrestrial and airborne laser scanning. In a first step, we apply a multiscale clustering and classification of local point set neighborhoods by considering their geometric shape. Corresponding feature values quantify the similarity to curve-like, surface-like, and solid-like shapes. For selecting and extracting surface features, we build a hierarchical trivariate B-spline representation of this surface feature
function. Surfaces are extracted with a variant of marching cubes (MC), providing an inner and outer shell that are merged into a single non-manifold surface component at the field’s ridges. By adapting the isovalue of the feature function the user may control surface topology and thus adapt the extracted features to the noise level of the underlying point cloud. User control and adaptive approximation make our method robust for noisy and complex point data.