Improvements on obstacle detection in the pantograph gauge due to the recognition of steady arms

The paper describes the latest improvements in the video-based obstacle detection system developed at Deutsche Bahn AG (Germany´s leading railway operator). The system automatically detects obstacles in the pantograph gauge at a distance of up to 70 m and retracts the pantograph before collision with an obstacle. Due to the multitude of similar-looking objects within an epipolar line, the algorithms developed gives false obstacle warnings in over 0.7% of the images. With recognition of steady arms, the reliability of the system is improved considerably. With a constant detection ratio, false obstacle warnings can be reduced by 28%.