A unified and convenient extrinsic calibration method for RGB camera, thermal camera, LiDAR and radar

• Main Author: Zhang, Fangwei
• Other Authors: Wang Dan Wei
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering
• Online Access: https://hdl.handle.net/10356/173518
• Institution: Nanyang Technological University

With the continuous advancement of autonomous driving technology, sensor ex trinsic calibration, as an indispensable part of it, has attracted significant at tention. The mostly common used sensors are 3D LiDAR, 3D radar, thermal camera and RGB camera. The main challenges faced in unified extrinsic cali bration include the following points: 1) the existing calibration target is hard to provide the common feature for four or more types of sensors due to the het erogeneity of sensors; 2) Even some methods that can realize unified calibration may still face limitations in terms of sensor positions or range of angles. This work is proposed to address these difficulties with the following contribu tions: 1) a spherical calibration target with the inner structure of eight trihedrals is proposed to provide the common feature; 2) the corresponding automatic fea ture extraction algorithms for sensors are developed sequentially; 3) the extrinsic parameters are regressed by min3D and min2D; 4) the results of calibration are evaluated qualitatively and quantitatively. The final error of extrinsic parame ters of different sensor pairs are listed as follows: 1) 0.06m for transformation from LiDAR to radar; 2) 0.84 pixels for transformation from LiDAR to ther mal camera; 3) 0.45 pixels for transformation from LiDAR to RGB camera; 4) 2.74 pixels for transformation from radar to thermal camera; 5)1.83 pixels for transformation from radar to RGB camera.