Formalising Expert Knowledge for Building Information Models: Automated Identification of Electrical Wiring from 3D Scans

New computational methods provide means to deduce semantic information from measurements, such as range scans and photographs of building interiors. In this paper, we showcase a method that allows to estimate elements that are not directly observable - ducts and power lines in walls. For this, we combine explicit information, which is deduced by algorithms from measured data, with implicit information that is publicly available: technical standards that restrict the placement of electrical power lines. We present a complete pipeline from measurements to a hypothesis of these power lines within walls. The approach is structured into the following steps: First, a coarse geometry is extracted from input measurements; i.e., the unstructured point cloud which was acquired by laser scanning is transformed into a simplistic building model. Then, visible endpoints of electrical appliances (e.g. sockets, switches) are detected from photos using machine learning techniques and a pre-trained classifier. Afterwards, positions of installation zones in walls are generated. Finally, a hypothesis of non-visible cable ducts is generated, under the assumption that (i) the real configuration obeys the rules of legal requirements and standards and (ii) the configuration connects all endpoints using a minimal amount of resources, i.e. cable length.