Relating images to BIM models for automated inventory

The aim of this project is to conceive an AI-based object recognition system to identify building components.

Today, the ever-increasing availability of digital Building information modeling (BIM) models make a valuable contribution to the maintenance phase in the life cycle of buildings. Additionally, the upcoming “digital twin” allows to complete the structural reality with a digital illustration that domesticates all the relevant information about the building. This digital illustration already forms the carrier for all digital processes of maintenance and operation and in the future will still provide a multiplicity of additional services (e.g., IoT, incident management, building control, user feedback).

The closer the structural reality and his digital twin are coupled to each other and the easier the access to the digital data from the current situation on site is, the more efficient, cost-effective and secure processes can be designed in the future. As of today, however, there exist no solutions to link real-world observations, such as images of building objects, with their BIM counterparts.

The proposed solution provides a key technology for this purpose. Under the title “On the fly”, the Computer Vision Laboratory (CVLAB) will develop an AI-based software system that uses artificial intelligence to recognize objects recorded on-site and assign them to their digital equivalents. This will provide a non-technical, user-friendly, cost-efficient and scalable coupling of real and digital, which opens up previously unimagined usage possibilities.

The eighteen-month project is carried out by the CVLAB in collaboration with Kaulquappe and Swiss Federal Railways. It is financed by Innosuisse.

Principal investigator Mathieu Salzmann
Period 2019-2021
Sponsor Innosuisse
External partners Kaulquappe AG, CFF SBB FFS
Laboratory CVLAB