Efficient Surface Reconstruction for Piecewise Smooth Objects

Abstract

In this report we present a new surface reconstruction technique from unstructured point clouds for piecewise smooth objects, such as scans of architectural and other man-made artifacts. The new technique operates in three conceptual steps: First, a set of basis functions is computed and a topology is established among these functions that respect sharp features using a RANSAC technique. Second, a linearized, statistically motivated optimization problem is solved employing this discretization. Lastly, an implicit function based meshing technique is employed to determine a clean, manifold mesh representation. The main benefit of our new proposal in comparison to previous work is its robustness and efficiency, which we examine by applying the algorithm to a set of synthetic and real-world benchmark data sets.