One-step method to calibrate a 3D LiDAR and a monocular camera

Abstract

Nowadays, autonomous mobile robots rely on heterogeneous sensors to enhance their perception capabilities, because of the complementary characteristics provided by different sensors. For example, cameras and Lighting Detection and Ranging sensors (LiDAR) are commonly used. To deeply fuse the information collected by different sensors, a precise extrinsic calibration is necessary, with which the transformation matrix (rotation and translation) between the sensor frames can be obtained. However, to calibrate a thermal camera and a sparse 3D LiDAR, the existing two-step method requires a visual camera to assist the process. In this thesis, a one-step method is proposed to calibrate a thermal camera and a sparse 3D LiDAR (Lighting Detection and Ranging). The proposed method completely removes the visual camera in the existing two-step method, while ensuring calibration accuracy. The complexity of the calibration process is greatly simplified and the calibration efficiency is improved. At the same time, a pre-processing step is proposed to calibrate a 2D rotating LiDAR and a monocular color camera, with the use of an off-the-shelf KITTI calibration toolbox. The 2D rotating LiDAR produces a very dense point cloud, which makes it impossible to obtain a correct result with the calibration toolbox. The method proves that after pre processing the original point cloud, the KITTI calibration toolbox is fully applicable to the calibration between the monocular camera and the 2D rotating LiDAR. The influence of different factors on the calibration results is also compared and discussed