Surface reconstruction of 3D point clouds based on greedy projection triangulation

This paper presents a preprocessing and surface reconstruction pipeline designed for point clouds in complex industrial environments. The proposed method starts with a density-adaptive voxel downsampling technique that adjusts voxel size according to local point density to reduce redundancy while preserving geometric features. Random Sample Consensus–based plane segmentation is then applied to extract dominant planar structures, followed by statistical outlier removal to suppress sparse and noisy points. For surface reconstruction, an improved Greedy Projection Triangulation algorithm is employed to incrementally generate a triangular mesh. Experiments on benchmark datasets are conducted to analyze reconstruction completeness, geometric detail preservation, and computational efficiency. The proposed pipeline is applicable to industrial point cloud processing tasks such as surface modeling, inspection, and reverse engineering.