A Multi-sensor Demarcation Method Oriented to Non-contact Train Obstacle Detection

In automatic train operation, non-contact train obstacle detection systems have replaced traditional manual sighting verification. One obstacle detection system usually combines multiple sensors including high-definition cameras, lasers, etc., and these sensors require regular demarcation and/or calibration. However, displacement of sensors would happen due to train operational vibration, and displaced sensors need to be figured out and second calibrated during the maintenance. Some of the current demarcation methods require special equipment and experienced maintenance workers; hence these methods have difficulties on implementation and are expensive. In response to the above issues, this article proposes a new method of automatic demarcation and calibration approach with a system of two cameras and one laser radar sensors. This method applies joint demarcation and calibration via definite cross-camera constraints; and it could ensure accurate external reference and parameters between two cameras and between one camera and laser radar senor. Besides, the method does not require definite correspondence between camera images and laser radar point cloud, thus the system could also be robust in open air environment. The test result shows that the erroneous calibration rate is remarkably reduced compared to the rate from the traditional calibration methods.