Simulation of pedestrian behaviour in traffic situations using risk-based A* pathfinding

Fatally injured vulnerable road users, especially pedestrians who collided with motorised vehicles, account for 31% of all recorded fatalities in urban traffic in the EU. Autonomous vehicles will improve this situation in the future, reducing the impact of the human factor in critical traffic situations. The development of autonomous driving functions requires simulation environments to train certain behaviours. Consequently, these simulations need well represented vulnerable road users. In this paper, an approach for accurate prediction of pedestrian behaviour at street crossings is presented. The suggested solution involves an agent-based model working with A* pathfinding and risk-based areas, validated by a drone dataset on German road crossings.