Filling Holes in Point Clouds

Laser scans of real objects produce data sets (point clouds) which may have holes, due to problems with visibility or with the optical properties of the surface. We describe a method for detecting and filling these holes. After detecting the boundary of the hole, we fit an algebraic surface patch to the neighbourhood and sample auxiliary points. The method is able to reproduce technically important surfaces, such as planes, cylinders, and spheres. Moreover, since it avoids the parameterization problem for scattered data fitting by parametric surfaces, it can be applied to holes with complicated topology.