Extrinsic calibration between multiple 3D LiDARs for autonomous robots

• Main Author: Lyu, Qiyang
• Other Authors: Wang Dan Wei
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/152597
• Institution: Nanyang Technological University

The autonomous vehicles and intelligent transportation are of vital importance in recent years. A single LiDAR only has limited field of view (FOV) and may have blind spot. Applying multiple sensors in one system can offset these drawbacks and further improve the system’s perceptual capability. Thus, a feasible method is to fuse the information from different LiDARs together. Calibration between multiple LiDARs are needed before fusing. The calibration process will figure out the spatial relationship between two sensors, which is rotation R and translation t. An accurate calibration is essential for a good fusion. Most of recent methods focus on short baseline condition. However, in intelligent transportation system in urban environment, LiDARs are commonly placed far away from each other, with large baseline, resulting in different perspective. Previous methods could not work under significant different perspective condition. A new calibration method is proposed in this thesis regardless of the restrictions of the baseline length. A sphere target is designed for the calibration, using the ICP method with known correspondence. The advantage of using a sphere is that the sphere is visible in all directions, and the sphere center keeps unchanged. With the usage of sphere, no prior assumption is needed for the calibration, except the necessary overlap FOV between multiple LiDARs. The stated method could work in both short baseline and long baseline condition. Simulations using sphere target are conducted. Translation and Rotation matrix between LiDARs are calculated using the proposed method, where the result turns out to be extremely accurate. The noise robustness test also shows impressive input error tolerance using the proposed method. The quantitative and qualitative test presents that the final error could be controlled under 0.01m and 0.1 degree under detection distance of 30 meters and noise with the standard deviation of 0.03m.