OBSTACLE DETECTION WITH A SEQUENCE OF ULTRA TELEPHOTO CAMERA IMAGES

In order to detect obstacles on railway tracks, this paper developed a forward monitoring system, which plays an important role in ensuring the safety train operation. This paper proposes a method of detecting obstacles with an image sequence taken by an ultra telephoto lens camera mounted on a train that monitors the status of the track for 600m (breaking distance) or over ahead of the train. First, railway track environment recognition is discussed. Particularly extraction of two parallel rails is a fundamental issue in applying image processing to railways. It is inevitable to limit the area where obstacles exist. A first detection algorithm is based on the analysis of optical flow, which characteristically occurs in those moving obstacles. Assuming that pixels corresponding to the same object have similar flow vectors, a connected region with similar flow vectors can be extracted. Then the paper focused on the stationary obstacle edge and the brightness contrast difference from the background. Compensation for the flow field caused by camera blur was performed, so that a vibration-proofing device would reduce blurs of the image and a wide dynamic range camera so as not be influenced by fluctuations such as outdoor lighting inconsistency. The experimental results using a real image sequence proves the effectiveness of the proposed method.