The Point Cloud Surface Reconstruction With Thin Layer Structure

In the media of digital geometry, the triangular grid and the point model are the two important ways to describe the three-dimensional discrete point cloud data. Along with hardware to obtain three-dimensional object data capabilities and the rapid development of computer software technology, the point model is becoming one of the main forms of3D media expression. Point model with the data structure is simple, convenient to access, especially suitable for the representation and processing of large-scale3D model, thus, geometry processing algorithms dealing with the point cloud are a hot research topic. In this paper, we study around three-dimensional reconstruction about the scattered point cloud and two-dimensional images.First of all, a survey is given to review the development of the point cloud geometry processing technology, and the research status in the current domestic and international. We also summarize the three-dimensional surface reconstruction and finite element mesh generation methods.Secondly, a mesh surface fitting method is proposed based on Laplacian deformation for reconstructing armor plate surfaces in this paper. Firstly, the proposed method finds sparse feature points on two dimensional images and their corresponding reconstructed3D points. A quadrilateral mesh is reconstructed using the sparse3D points as control points. It reconstructs the3D surfaces by the deformation method of the least-squares mesh. The core of the method is the generation of the connectivity mesh and finding the correspondences between the3D points and the vertices of the connectivity mesh, which can be easily solved using the information of the original2D images. The proposed approach is fast and simple without any additional meshing and parameterizations. The paper demonstrates the effectiveness of the approach with several examples.Finally, this paper presents a structure with a thin layer scanning point cloud data grid reconstruction method. In order to overcome the common noise and outliers in the scanning point cloud, first, we use the local optimal projection (LOP) operator to smooth the original data; Secondly, the geometry for the model layer by the method of spectral clustering, the point cloud data is sliced; Finally, we treat respectively both of the slices with method of estimation and consistent orientation. Lots of experimental results show that the method not only can effectively handle noise, outliers, such as scanned point cloud non-uniform distribution of the common problems, but also can reconstruct the thin structure of the grid with the correct model.