Obstacle recognition from forward view images from trams

Abstract

This research is concerned with the safe operation of tram systems. We construct a warning system to the tram driver that recognizes obstacles from the forward view images from the tram cab. For this realization, front view images are processed, the obstacles in the image are recognized, and are judged either safe or dangerous. This methods characteristic is to use a moving mono camera. At first, the danger area is limited by detecting the track. Static obstacles are detected by the variation of the intensity histogram between the tracks. Moving objects are detected by frame difference considering the trams velocity in 3 dimensions. To recognize the obstacles the following characteristic parameters are adopted: - Argument of diagonal in outside square of the detected obstacle - Obstacles width By the recognition, obstacles in the images are classified into four categories: Man, Bicycle, Car and The other. Collision or avoidance probability is judged by velocity vectors of the tram and obstacle. If the expected case of collision for Man, Bicycle and Car is danger, then the tram driver is warned by the system. We inspect the reliability of the image recognition in 13 scenes 14 images; the rate of correct judgment reaches 85 and collision judgment reaches 92.