Improving safety of level crossings by detecting hazard situations using video based processing

Road and level crossing safety become a priority issue for the domain of intelligent transportation systems in recent years. This paper presents a video based approach for detecting and evaluating dangerous situations induced by users (pedestrians, vehicle drivers, unattended objects) in level crossing environments. The approach starts by detecting and tracking objects shot in the level crossing area thanks to a video sensor. Then, a Hidden Markov Model is developed in order to recognize ideal trajectories of the detected objects during their tracking. The level of risk for each identified hazard scenario is estimated instantly by using Demptster-Shafer data fusion technique. Three hazard scenarios are tested and evaluated with different real video image sequences: presence of obstacles in the level crossing, presence of stopped vehicles lines, vehicle zigzagging between two closed half-barriers).