Extrinsic parameter calibration methods of sensors present in a robot tractor

To achieve large-scale deployment of autonomous agricultural machines, a reliable and accurate perception system must be developed. Often, autonomous machines use and require data from several sensors to work optimally and safely. Therefore, accurate extrinsic parameter calibration between various sensors is one of the first prerequisites. This paper equips an actual robot tractor with two LiDAR sensors, a stereo camera, and a GNSS/IMU unit and it investigates different extrinsic calibration methods between them in the agricultural environment. The extrinsic calibration method between the LiDAR sensors utilizes extracted planar structures from point clouds. In the case of the LiDAR and the GNSS/IMU unit, two calibration methods are developed. The first method utilizes LiDAR point cloud features, whereas the second method uses sensor motion estimates. In turn, the LiDARs and the camera are extrinsically calibrated using a traditional checkerboard method. The results with the actual robot tractor show that the methods achieve correct and consistent calibration results in the agricultural environment. The optimization functionality of the LiDAR-to-LiDAR calibration method was validated using simulation. In addition, the actual results are validated by cross-validation by calculating extrinsic parameters between LiDAR sensors using the other methods mentioned above. The average standard deviation of the results is 0.3189∘ for rotation and 0.0491 m for translation parameters. In addition, a visual examination of the results strengthens this conclusion.