As a part of the digital revolution of railway systems, an autonomous driving train will use a complete and precise map of railway infrastructure to conduct operational actions. Nevertheless, the full autonomy of trains depends on the safety decisions management capacity both on-board and track-side. These decisions must be refined into safety requirements in order to continuously check the consistency between the perceived infrastructure and safety related properties. However, traditionally, the integration of safety analysis requires the intervention of human agent skills. This may be error-prone and in interference with the embedded aspect of the train map. In this paper, we propose a model-based approach to match between safety concepts expressed as an ontology, a derived safety model and a safety-extended railway infrastructure map model for autonomous trains. This approach is validated by railway safety case studies for autonomous train map. The integration of this model-based safety solution from the early stages of the map system design improves the safety decisions management process.