An Efficient Method to Mesh Point Cloud

This paper presents a new reverse engineering method for creating 3D mesh models, which approximate an unorganized noisy point set without orientation information. The new method computes sample points by the extended moving least squares method in adaptive octree cell. The octree subdivision is decided by weighted covariance matrix. Then the points are connected by intersections of supported spheres. Further, the triangular meshes are refined to remove non-manifold parts and holes. The new algorithm allows us to construct mesh models from very large point set quickly