Research On CAD Modeling Based On Measured Data

CAD modeling based on measured data is an important part of CAD/CAM. It has been applied in many fields such as manufacture, advertisement and science visualization. Many precious research achievements have been made in this field, but there are still some key and difficult questions waiting for the solutions. This dissertation is concerned with four important techniques of CAD modeling based on measured data. These techniques are mesh reconstruction, mesh repairing, mesh segmentation, and quadrilateral domain extraction from the mesh model. The main contributions are as follows:A novel mesh reconstruction algorithm based on local smoothness is proposed. The algorithm firstly computes the neighborhood smoothness of each point and constructs an initial seed face in the smooth region; then the active edge from the smooth region is chosen priority to grow; finally, a manifold mesh is generated by a post process. In each growing, the neighborhood smoothness is introduced to select the best candidate point for constructing a new triangular face. The algorithm is simple and robust, and can obtain satisfactory result for the thin-plane measured data and reconstruct the features of the underlying-object naturally. Based on geometrical distritbution of the point set, a novel algorithm for the reconstruction of surfaces from unorganized sample points is presented. Points are classified based on their geometrical properties. A neighborhood is generated for each point and adaptively refined based on length and angular distributions. A mesh is reconstructed starting from smooth regions to highly curved regions. The algorithm is robust and effective; and can procude high quality meshes while preserving features of the geometrical shapes.To repair the complex holes in triangular meshes, a novel algorithm based on the concept of edge-expansion is proposed. A hole boundary is projected onto the least square plane fitted to the hole boundary and the intersecting edges are recorded. The edge-expansion algorithm is used for each intersecting edge to obtain several triangular faces, and then the complex hole is divided into several sub-holes. The complex hole-splitting process mentioned above is incrementally employed for each sub-hole until all sub-holes are divided into simple ones, and then each divided simple hole is repaired with planar triangulation method. The algorithm can work well for a variety of complex holes in the triangular meshes, and better preserve the detailed features of the original triangular mesh. An algorithm for filling holes based on vertex clustering is presented. It firstly selects the relatively flat point, and expands along the hole boundary until the max deviation of polynomial surface with degree two fitted to the expanded vertices is larger than the prescribed threshold; then two new holes are generated when the geometrical shapes of them are reasonable. The hole splitting process above-mentioned is incrementally employed for the unexpanded holes, and then fills each sub-hole. The presented algorithm is robust, and can better preserve the geometrical features of the mesh.To segment triangular meshes into several meaningful part components, a novel algorithm based on geometrical structure signature is presented. The Poisson shape signature of each face in the model is firstly computed. The mean-shift algorithm is employed to cluster the Poisson shape signature of each face. Based on the Poisson shape signature, the core part component is extracted, and the boundary between part components is refined. The presented mesh segmentation algorithm can work well for the triangular meshes with part component structure, and is independent of the part rigid-transformation.A novel algorithm to extract quadrilateral domains on the mesh is presented. It firstly segments the mesh into flat and disk-shaped sub-meshes with quasi-convex boundaries; then extract quadrilateral domain for each sub-mesh. The algorithm can automatically capture the prominent geometrical features of the model and extract the quadrilateral domains which can reflect the intrinsic direction of the model. Also the user can edit the mesh segmentation results to obtain a high quality quadrilateral domain extraction. For the mesh with component structure, we extend our algorithm. It firstly segments the mesh into meaningful components, and then extracts the quadrilateral domains.