Frontal Obstacle Detection Using Background Subtraction and Frame Registration

抄録

Systems such as Automatic Train Protection and moving block sections help prevent trains colliding, however collisions with unexpected obstacles in front of a train can only be avoided if seen by the driver. In an effort to reduce the possibility of this type of collision and to improve passenger safety, an obstacle detection method has been proposed using a monocular camera and image processing. The proposed method can detect obstacles by comparing live images from the camera with images obtained by other trains operating earlier along the same route. The difference between the two sets of images are defined as obstacles. The performance of the method was verified by conducting experiments using rolling stock and imitation obstacles.