The development of a technique for 3D complex surface reconstruction from unorganized point cloud

In this paper, we propose a new neural network based on our two-level adaptive hierarchical clustering algorithm. The algorithm is to manage unorganized points, so that the triangular mesh models can be correctly obtained by applying the triangular mesh creation algorithm. We also develop adaptive self-flipping triangle edges to improve triangular mesh structure. Only one parameter, the maximal edge length of triangle, is needed in the neural network. The proposed two-level consists of the first level for clustering the cloud of points that has same order of the maximal edge length into a same cluster and the second level for generating triangular surface model or drape surfaces over the points of the same cluster. The normal vector for the generated triangular 3D surface model can be obtained from the second level. This helps to generate the STL file or stereolithography format. From the experimental results it can be shown that the proposed method is very effective for clustering unorganized point clouds for generating a triangular mesh of complex surfaces.