Video-based obstacle tracking for automatic train navigation

For automatic train navigation, we have to consider obstacles existing on railway and moving toward railway in order to determine control states such as sudden break, gradual break, alarm, and so on. This paper proposes a method of video-based obstacle tracking for automatic train navigation. In this method, the tolerance for safe running is firstly determined due to the train speed, the train would be suddenly stopped, gradually stopped, horned according to the location of static obstacle. In case of dynamic obstacle that is moving to train running zone, trajectory obstacle is estimated, and train control state is prior consider in order to keep train from crashing. In this paper, we employ perspective transform to obtain each zone on the image plane in advance, and detect both static and dynamic obstacles on the image plane while the train running. The experiments are performed on video, the results reveal that by the proposed method, all zones are dynamically calculated due to the train speed, and both dynamic and static obstacles are correctly detected.