3D Building Modelling from ALS Point Clouds by Delaunay Triangulation and Graph Theories

Abstract

As an important task in 3D building reconstruction, 3D roof reconstruction attracts increasing attention. Existing methods have the issue of additional errors caused by multi-step processes, resulting in relatively high uncertainty and potentially affecting the reconstruction accuracy. Deep learning-based methods to achieve the direct extraction of roof vertices and edges for reconstructing 3D roof structures tackle the issue of additional errors while leading to another problem: the reliance on large labeled datasets for training. In this study, a fully rule-based method is proposed to achieve automatic 3D roof reconstruction. Roof vertices with their edges parallel to the x-y plane are first extracted from the point clouds, and subsequently, the left roof edges connecting vertices are inferred by using Delaunay triangulation. Finally, the 3D roof structure is reconstructed by using graph analysis based on the information of roof vertices and edges. This method simplifies the process of 3D roof reconstruction, and the experiment results demonstrate that the proposed method can effectively reconstruct 3D roof structures from point clouds.