Repairing 3D models obtained from range sensors

Matlab toolbox developed to:

• Automated Hole Identification: The algorithm automatically identifies "badly sensed" areas or holes in 3D polygonal meshes by detecting boundary edges that are not shared by two triangles (non-manifold edges).
• Intelligent Mesh Portion Extraction: Instead of processing the entire model, the code extracts a specific "zone of interest" around each hole. It calculates the optimal size of this area based on the mesh resolution and the average length of the edges.
• Curvature-Adaptive Selection (Normal Diffusion): To avoid projection errors in complex areas, the method includes a normal diffusion algorithm. It monitors changes in surface curvature and limits the growth of the processed area if the normals change too abruptly.
• 3D-to-2D Optimized Projection: The system projects 3D mesh data onto an automatically computed reference plane that maximizes the projected area of the hole. This converts 3D coordinates into a 2D range image where pixel values represent height.
• Handling of Complex/Self-Intersecting Holes: The code can detect if a hole's boundary self-intersects when projected. In such cases, it automatically subdivides the complex hole into simple sub-polygons to process them independently.
• Inverse Transformation & Triangulation: After the 2D image is repaired, the code performs an inverse 2D-to-3D transformation. It then applies a Delaunay-based triangulation (specifically 2.5D) to create a new mesh patch that is seamlessly integrated into the original model.
• Robustness to Non-Sensed Zones: The framework is specifically designed to recover data lost due to occlusions or inaccessible crevices during the scanning process with range sensors.