Integrated Quality Mesh Generation for Poisson Surface Reconstruction in HPC Applications

Finite Element Analysis (FEA) is a numerical method for solving engineering problem. Implementations of FEA are common high performance computing (HPC) applications. As a typical application of FEA, surface reconstruction of large model, proves to be a computationally expensive task as well. Poisson Surface Reconstruction, a cutting-edge surface reconstruction algorithm, is a commonly used method for efficiently solving linear systems and it creates watertight surfaces from oriented point sets. Poisson Surface Reconstruction leverages an octree-based Marching Cubes (MC) method for isosurface extraction. Hierarchical octree structure avoids unnecessary cells visiting and prevents cracks arising. In this paper, we integrate several quality improved methods with MC and coordinate unrelated components, obtaining a better quality mesh implementation. We improve mesh quality based on an extended lookup table and modify the connectivity of some fundamental patterns in MC, which effectively remove the reconstruction holes, thus improving overall surface quality. As for the relative value between each vertex and the average isovalue, the extended table explicitly differentiates between “strictly larger” and “equal to”. Newly introduced patterns in MC statistically prevent poor quality triangles production. Moreover, a decision making algorithm is proposed to eliminate ambiguity problems. We adapt SnapMC algorithm to avoid non manifold triangles to a certain extent. Comparisons with traditional Poisson algorithm and Smooth Signed Distance (SSD) highlight the capability in quality mesh generation and efficacy in handling high computational demand.