Scenario-Based Validation of Automated Train Systems Using a 3D Virtual Railway Environment

The latest advances in Deep Learning allow for increasing the automation and safety of driving systems. However, the shortage of openly available railway datasets makes it difficult to validate automated train systems. In this paper, we present a methodology to build 3D virtual railway environments for a scenario-based validation of perception systems on simulated data. The virtual environment is designed based on an ontology for railway environments following the data model of ASAM OpenLABEL. The environment is then employed to generate scenario-based synthetic data and validate perception systems in different conditions. Finally, we showcase our methodology with a practical use case focused on the validation of a rail-track detection model in different domains.