Point Cloud Surface Reconstruction

The field studies how to infer a usable surface from 3D points collected by sensors, scans, or multiple images, enabling tasks such as measurement, mapping, and object modeling. Common approaches include implicit signed or unsigned distance fields, learned and generative shape models, and explicit mesh methods such as Poisson reconstruction and ball pivoting; some also use visibility, normals, or temporal observations as constraints. Methods are evaluated on object and scene benchmarks, as well as LiDAR, photogrammetry, and dynamic-scanning settings, using measures of geometric accuracy, completeness, smoothness, and runtime. A central tension is reconstructing fine or thin geometry from sparse, noisy, and occluded data without sacrificing robustness, scalability, or computational efficiency.