Analysing Railway Accidents: A Statistical Approach to Evaluating Human Performance in Obstacle Detection

This paper evaluates human driver performance in obstacle detection on railway systems by statistically analysing accident data.Given the challenge of obtaining nonaccident data, the study estimates obstacle frequency on tracks by examining accident rates under low visibility conditions (night and curves), where drivers cannot see obstacles in time.Using data from a major European railway operator, the study finds that human drivers can avoid around 28% of collisions in good visibility conditions, but only about 12% when all conditions are considered.Autonomous trains need to meet at least an equivalent performance level.This research aims to support the safety certification of autonomous train systems by offering a benchmark of human performance in obstacle detection, emphasising the need for comprehensive data and rigorous validation of hypotheses regarding driver reactions and environmental conditions.