An autonomous robotic platform for automatic extraction of detailed semantic models of buildings

This paper presents an integrated system that automatically provides detailed as-is semantic 3D models of buildings. The system is able to explore and reconstruct large scenes at a high level of detail, passing through five semantic levels, finally generating a detailed semantic model of the building. Our autonomous scanning platform collects raw data regarding the scene. At the first level of modelling, our autonomous scanning platform collects data regarding the scene and generates a point cloud that is later structured in a semantic point cloud model containing indoor, clutter and outlier point clouds. The second and third levels of semantic models consist of a simple B-rep representation and a model of basic building components, which includes the walls, ceiling, floor and columns, as well as their topology. Openings are then added, thus yielding our fourth semantic model. Finally, small components in buildings, such as sockets, switches, lights and others are recognised, resulting in the fifth semantic model. This approach has been tested on real data of building floors using our Mobile Platform for Autonomous Digitization (MoPAD). To the authors' knowledge this is the first work that, after obtaining 3D data with an autonomous mobile scanning platform, achieves such detailed modelling of building interiors. The performance of the method has been assessed quantitatively against ground truth on simulated and real environments. Two videos are available at the Supplementary material of this paper.