Optimizing B-Spline Surface Reconstruction for Sharp Feature Preservation

Methods of surface reconstruction from 3D point clouds have received much attention in recent years due to their vast array of applications and the increasing supply of accurate 3D data. Providing smoothness, local modification, and robustness to noise, the B-spline surface fitting is one of the most popular of such methods. However, a problem encountered when using B-spline surface reconstruction is the representation of sharp features: corners and edges tend to be smoothed out. We propose an approach to sharp feature preservation which relies on curvature analysis of the B-spline surface. B-spline patches that have high curvature and are surrounded by patches with low curvature are identified as those representing sharp features. The location of sharp features is then determined through interpolation from low-curvature patches surrounding the identified patches. Finally, these features are preserved through repeated addition of points to the point cloud. We evaluate our sharp feature preservation algorithm at varying levels of noise, demonstrating its high accuracy at low noise and moderate robustness as the noise increases.