Research On Surface Reconstruction Based On Delaunay Refinement From Scattered Point Clouds

Reverse engineering, as an important design and manufacture technique, is significant to shorten the period of product design and manufacture in CAD/C AM. It can be widely applied in automobile, archaeology, aerospace, weaponry and other fields. With the rapid development of computer technology and digital measurement technology, the means to obtain point clouds are more abundant, faster and more convenient. How to reconstruct surface with scattered point clouds has become a bottleneck in the development of reverse engineering. This paper focus on surface reconstruction of different types of point clouds by using Delaunay refinement. The specific contents are as follows:(1) Delaunay-based surface reconstruction method requires that the point cloud satisfy r-sample rules. The method is not suitable for massive point cloud. A non-uniform sampling algorithm was proposed, which preserve boundary points well and down-sample point cloud adaptively according to different surface features. Experiments show that the simplified point cloud can preserve boundary well and is suitable for Delaunay-based surface reconstruction method.(2) A new surface reconstruction algorithm based on Delaunay refinement for noisy point cloud was proposed. Firstly, according to moving least square algorithm, robust estimation theory was introd uced to algebraic sphere fitting so as to fast and robustly approximate the local surface. Secondly, the balls surrounding surface were divided by AABB-tree so that the balls intersected with any given segment could be quickly founded. The segment-surface intersection could be calculated by parallel computation. Finally, the surface was reconstructed by Delaunay refinement. Experiments show that the new algorithm can reconstruct surface with high precision and good aspect ratio.(3) A new feature-preserving surface reconstruction algorithm for noisy point cloud with sharp features was proposed. Firstly, the initial feature points were detected by Voronoi covariance matrix method and clustered using feature line direction. Secondly, each feature point was moved to its neighbors’ projecting center along its feature direction. The feature points were uniform down sampled according to the moved distance. Thirdly, the feature line was reconstructed using NNCrust algorithm. The corners were repaired with the constra int of nearby feature lines. Finally, the surface was meshed by Delaunay refinement using protecting ball. Experiments show that the new algorithm is robust and can preserve sharp features well.(4) The surface reconstruction algorithm based on Voronoi covaria nce matrix was improved for noisy point cloud with small amount outliers. The differential equation of implicit function was constructed such that its gradient is most aligned with the principal axes of the Voronoi covariance tensor filed. As a result, the surface reconstruction problem is transformed into a generalized eigenvalue problem. The differential equation is solved in discrete exterior form. The probability measure theory is introduced to define offset surface when discretizing point cloud space which makes the algorithm more robust to outliers. Experiments show that the algorithm can robustly reconstruction surface with good aspect ratio without normals.