3D Surface Extraction Using Incremental Tetrahedra Carving

We propose a fully incremental and yet globally optimal surface extraction method for updating the 3D surfaces when new images, camera poses and 3D points are additionally registered into the existing 3D model. We extend the tetrahedra-carving-based surface extraction algorithm to the incremental fashion by efficiently detecting ray-tetrahedra intersections and formatting the graph to solve with the dynamic graph cut. The proposed method can result the surfaces identical to the one running from scratch, i.e. the global optimality of the extracted surface is guaranteed while improving the efficiency with fully incremental procedures. We compare the proposed method with state-of-the-art baseline methods and finally demonstrate the surface extraction of fairly large objects.