Abstract
Surface reconstruction problem is a classical problem which ?nds wide applications in di?erent disciplines, such as science, engineering and arts. For well-sampled point set from a smooth surface, there have been many algo-rithms to correctly reconstruct a triangular mesh from the point set. However, surface models in reality usually contains some sharp features, such as crease lines, corners and apices where C-1 continuity is not guaranteed onthe surface. The acquisition process of point clouds are usually not adjusted to obtain the sample points sitting right on the sharp features of the given model. In fact, almost none of the sample points acquired lie exactly on such sharp features. Therefore, the information of sharp features of the original surface model are lost during this sample acquisition process. Our goal is to propose a new and e?ective method that not only gives good reconstructionto the smooth areas of the given surface, but also recovers the sharp features originally existed in the source surface model provided that a reasonably nice sampling point cloud is given. We present an algorithm that elegantly make use of the polar normals obtained from the co-cone algorithm and the triangles on the smoother areas near the sharp features to estimate the positions of the inserted pointsapproximating locally sharp features. We have also conducted some preliminaries experimentation on some surface models with di?erent kinds of sharp features, and use the piecewise sharp feature groups to estimate corner orapex features. Also, the identifying of group of sharp features can be a great information to handle the noised model, by constraining the de-noising process to the surface vertices without crossing sharp features. We ?nd that our algorithm signi?cantly improves the quality of the reconstructed surfaces in most cases.