Studies On Mesh Reconstruction And Feature Extraction

In computer graphics, digital geometry processing, CAD modeling, digital entertainment and other fields, it always needs to digitalize the model that obtained from 3D scanner. One model is point cloud model; it is a geometric model using a point as a primitive and has many excellent properties such as simple structure, compact in space and efficient in representing large scale geometry models with rich details. However, the point cloud model can not generally be applied directly, designers use different methods to reconstruct surface mesh models for different purposes; this approach is also known as reverse engineering. Another model is mesh model; but usually the 3D scanning system can only obtain one part data of the detected data, In order to obtain the complete data information, designer commonly used multi-angle, that is, from various angles to detect and scan the mode. Since the coordinate system is different in various measurement perspectives, registration must be adopted for the multi-angel meshes measured to convert them into the same coordinate system; so designer can get the entire shape information of the surface through mesh integration. As triangular mesh with many advantages is the mainstream express way in 3D applications domain now,so feature extraction based on mesh is also important for research and practical application.This thesis focuses on point cloud and mesh model as the research object, and have discussed the two ways to surface reconstruction, namely, point cloud mesh reconstruction and mesh integration. The handling way of these two methods are different, and have different application area. In addition, focus on the studies of the discrete mesh surface curvature calculation and feature extraction. This research is divided into the following three parts:(1) Presented one novel mesh reconstruction algorithm from point cloud. The proposed algorithm is based on geometric and shape feature of point cloud. First, points are classified according to their geometric distribution, and then points are classified according to the local shape feature. Then mesh reconstruction is done on different type of points. Finally, a post-processing is performed to repair the topological and geometric errors. Experimental results show that our algorithm is robust and effective, high quality meshes can be created while geometric and shape features are well preserved.(2) Presented one modified registration algorithm for scanned mesh model, and one mesh integration algorithm based on mesh reconstruction according to the registration algorithm is advanced. Firstly, the overlap region of the two meshes is detected. Secondly, the redundant triangles in the overlap region are deleted, and then resample the point cloud data of overlap region. Thirdly, a surface mesh is obtained through mesh reconstruction on that data. Finally, the entire surface is obtained by stitching with two old meshes on the boundary. Experimental results show that the algorithm is effectively for mesh integration, it is faster and robust.(3) Studied on and implemented the several discrete mesh surface curvature calculation method, and advantages and disadvantages of various methods were analyzed; and research on the feature extraction algorithms based on curvature calculation, the pros and cons of various algorithms and different parameters on the impact of feature extraction are analyzed.