Semantics-guided spatial data generation for complex large-scale indoor map from 3D colorized point clouds

Abstract. With the popularization of the concept of smart cities and the development of indoor positioning and navigation services, as well as the increasing complexity of building structures with the continuous advancement of urbanization, automatic mapping of large-scale indoor spatial data has become the fundamental work for subsequent real-life 3D applications. Therefore, this paper develops a semantics-guided generation method of indoor spatial data for mapping indoor spaces, where the roles of semantics are investigated for the subdivision and reconstruction of indoor spaces. It consists of the following four parts: (1) Semantic segmentation of 3D indoor scene; (2) Storey segmentation using semantics-enhanced height histogram; (3) Semantics-guided room segmentation based on building physical structures; (4) Room-wise boundary optimization using semantics-aware Recursive Search. Both quantitative and qualitative experiments are conducted on two public benchmark datasets: Stanford Large-Scale 3D Indoor Spaces (S3DIS) dataset and Matterport3D dataset. The results demonstrated that our method is capable of reconstructing complicated large-scale indoor scenes with higher robustness and reliability, outperforming existing state-of-the-art algorithms.