Developing a Comprehensive 3D Point Cloud Dataset for Construction Projects

3D reconstruction has become a common method to obtain digital representations of the built environment. Meanwhile, Scan-to-BIM technologies are investigated to extract semantic information from raw point clouds data. To improve the accuracy and reduce the time in object segmentation, Deep Learning (DL) techniques have been proposed to process 3D point clouds and proved as a promising approach in unmanned-vehicle driving. However, due to the little available dataset specified for construction projects, implementing deep learning in Scan-to-BIM applications is limited. Therefore, a point cloud dataset including the common scenes in construction is urgently desired. This paper is a part of a research project aiming to develop a comprehensive 3D point clouds dataset integrating the Red/Green/Blue (RGB), XYZ, intensity, and thermal data collected using digital cameras, terrestrial laser scanners, and infrared cameras. The collected data were first transformed into panorama (PAN) form to ensure enough and flexible field of view (FoV). Then, a registration method is proposed to fuse these data from different sources and to establish an integrated dataset. Two case studies were carried out in both indoor and outdoor environment to investigate the feasibility of the proposed methodology. The preliminary results show that feature-based registration method provides a reliable alignment between different data.