Connectivity detection for automatic construction of building geometric digital twins

Existing buildings' geometry digitisation using Point Cloud Datasets is important. This is evident given the lack of digital as-designed models for older buildings, which can serve as a reference for as-is conditions modelling, and their unreliability for newer buildings. However, detecting the connectivity between entities for effective construction of a geometric Digital Twin from Point Cloud Datasets is an unsolved problem. This work defines a new problem, connectivity detection in buildings from Point Cloud Datasets. To solve it, we define a surface topology graph that represents relationships between labelled surfaces and propose a deep-geometric-neural-network-based framework to reconstruct the graph. Extensive experiments on the S3DIS dataset demonstrate that our method achieves high performance in relation detection. The practical application for structural object detection further validates the effectiveness of the proposed approach and highlights the value of relations. The findings of this study contribute to the advancement of the construction of Digital Twins, facilitating the efficient analysis and reasoning of objects, spaces, and entire buildings within the Digital Twin framework.