3D surface defect detection and measurement using point cloud unwrapping via dimension reduction

Abstract

In building restoration, detecting 3D surface defects via laser-scanned point clouds faces challenges due to high-dimensional data complexity in visualization, computation, and interpretation. To address inefficiencies in surface defect detection from point clouds, this paper introduces a method based on point cloud unwrapping for building components. The methodology reconstructs point clouds into 3D meshes, unwraps them into 2D UV maps, converts the maps into color images for semantic segmentation, and uses bidirectional mapping for 3D defect localization and quantification. Experiments show the method achieves 20 % higher accuracy than direct 3D segmentation, with precise defect localization and geometric measurements. The findings demonstrate the effectiveness of integrating point cloud processing with computer vision techniques, which enhances detection accuracy while reducing dependency on manual inspection. Future work will focus on automating point cloud segmentation, optimizing complex geometry handling, and extending validation to diverse building materials.