An Automatic Hypothesis of Electrical Lines from Range Scans and Photographs

Building information modeling (BIM) with high level of detail and semantic information on buildings throughout their lifetime are getting more and more important for stakeholders in the building domain. Currently, such models are not yet present for the majority of today’s building stock. With increasing speed and precision of laser scans or photogrammetry, geometric data can be acquired at reasonable costs. Unfortunately, these data are unstructured and do not provide high-level semantic information, which stakeholder require for non-trivial workflows. A current research topic are methods to extract non-visible structures from visible geometric entities. This work uses domain specific geometric and semantic constraints to automatically deduce information that is not directly observable in architectural objects: electrical power supply lines. It utilizes as-built BIM data from scans of indoor spaces in order to provide a hypothesis of paths of electrical lines. The system assumes that legal requirements and standards exist for defining the placement of power supply lines. This prior knowledge is formalized in a set of rules, using a 2D shape grammar that yields installation zones for a given room. Observable endpoints (sockets and switches) are detected in indoor scenes of buildings using methods from computer vision. The information from the reconstructed BIM model, as well as the detections and the generated installation zones are combined in a graph that represents all likely paths the power lines could take. Using this graph and a discrete optimization approach, the subgraph is generated that corresponds to a probable hypothesis. Our approach has been tested against synthetic and measured data and shows promising first results. Application possibilities include generation of a probable wiring for as-built / optically acquired building model, or suggesting cable ducts for a building reorganization or during planning of a new building.