Proceedings of the

The 33rd European Safety and Reliability Conference (ESREL 2023)
3 – 8 September 2023, Southampton, UK

Creating 3D Models of Bridges Using Different Data Sources and Machine Learning Methods

Kwasi Nyarko Poku-Agyemanga, Maximilian Kellnerb, Annette Schmittc and Alexander Reitererd

University of Freiburg, Department of Sustainable Systems Engineering INATECH, Germany; Fraunhofer Institute for Physical Measurement Techniques IPM, Germany.


In today's world, aging and worn bridges pose an increasing risk to transportation infrastructure. In the worst case, old, poorly maintained bridges can collapse at any time. But complex and expensive maintenance work on the bridges causes traffic jams, which can lead to accidents or delivery problems. Therefore, bridges require intelligent and individual maintenance, which leads to a higher demand for documentation. One way to facilitate documentation is Building Information Modeling (BIM), which is based on a 3D model of the construction. For most of the German bridges no 3D data is available. So, it is necessary to create a 3D model as a base for the BIM by Scan-to-BIM processes. The 3D data for this process can come from a wide variety of sources like laser scanning, photogrammetry or analog 2D plans. A concept for automated 3D modelling with data from diverse sources and machine learning methods is presented. Point clouds of the bridges captured with cameras and/or laser scanners and 2D plans are used as data base for the 3D model, which is created by machine learning methods from the fused point clouds by calculating surfaces. The resulting model can be used for BIM and AR/VR applications.

Keywords: Machine learning, Point cloud, 3D reconstruction, Deep learning, Bridge.

Download PDF