Surface Reconstruction from Unorganized Point Cloud Data Via Progressive Local Mesh Matching

Download or read book Surface Reconstruction from Unorganized Point Cloud Data Via Progressive Local Mesh Matching written by Ji Ma and published by . This book was released on 2011 with total page 272 pages. Available in PDF, EPUB and Kindle. Book excerpt: Nowadays point cloud (a set of dense discrete points) has become an emerging data format to represent 3D surface geometry due to the increasing application of 3D laser scanning systems. Converting such a discrete point representation into a continuous surface representation is known as surface reconstruction. Many computer-aided design and inspection applications demand an accurately reconstructed surface corresponding to a watertight triangle mesh passing through the scanned point cloud data. Automatic reconstruction of a watertight triangle mesh with correctly represented sharp features remains an open issue in surface reconstruction research. This thesis presents an integrated triangle mesh processing framework for surface reconstruction based on Delaunay triangulation. It features an innovative multi-level inheritance priority queuing mechanism for seeking and updating the optimum local manifold mesh at each data point. The proposed algorithms aim at generating a watertight triangle mesh interpolating all the input points data when all the fully matched local manifold meshes (umbrellas) are found. Compared to existing reconstruction algorithms, the proposed algorithms can automatically reconstruct watertight interpolation triangle mesh without additional hole-filling or manifold post-processing. The resulting surface can effectively recover the sharp features in the scanned physical object and capture their correct topology and geometric shapes reliably. The main Umbrella Facet Matching (UFM) algorithm and its two extended algorithms are documented in detail in the thesis. The UFM algorithm accomplishes and implements the core surface reconstruction framework based on a multi-level inheritance priority queuing mechanism according to the progressive matching results of local meshes. The first extended algorithm presents a new normal vector combinatorial estimation method for point cloud data depending on local mesh matching results, which is benefit to sharp features reconstruction. The second extended algorithm addresses the sharp-feature preservation issue in surface reconstruction by the proposed normal vector cone (NVC) filtering. The effectiveness of these algorithms has been demonstrated using both simulated and real-world point cloud data sets. For each algorithm, multiple case studies are performed and analyzed to validate its performance.