A Sensor Calibration Method Based on Rail Detection

In recent years, environment perception systems have been developed to achieve autonomous operation of existing railways. To make the system reliable, fusing multiple sensor data is effective, and the sensors have to be calibrated for the sensor fusion. In this study, we propose a method to calibrate a LiDAR and a monocular visible light camera using rail detection results. By utilizing existing rails as targets, our proposed method does not require any new target installations and can carry out calibration before the train starts running. Considering some of the rail shape features, rails can be robustly detected using LiDAR data. With the camera, 2D rails are detected by referring to an existing method and converted to 3D rails. Extrinsic parameters of a sensor are calculated by using the point correspondence of the detected rails in different coordinate systems. Experiments showed that our method reduced the object location error due to inappropriate extrinsic parameters, and the estimated parameters were more accurate than another known method in use.