QUIETERRAIL D4.1 List of use cases and optimisation criteria QUIETERRAIL aims to support the development of quieter...
Deliverables: Results Published in June 2026
read more
Monitoring rail-corridor water from onboard cameras using foundation-model patch descriptors
Output type:
The IEEE International Conference on Intelligent Transportation Systems (ITSC) 2026
This study focuses on detecting water near railway tracks using onboard train camera videos. It uses a pretrained AI model with very little training data (only 5 labeled images) to identify and segment water areas. The results show high accuracy, though performance decreases for small or partially hidden water zones. Overall, the approach demonstrates that efficient, scalable monitoring of flood risks along tracks is possible, helping improve railway safety and maintenance planning.