• 机械基础工程 •

高鲁棒性两阶段激光雷达惯性测量单元外参在线标定算法

1. 1.华中科技大学能源与动力工程学院，武汉，430074
2.驭势科技有限公司基础平台研发部，北京，102400
3.华中科技大学机械科学与工程学院，武汉，430074
• 出版日期:2022-12-25 发布日期:2023-01-10
• 通讯作者: 张捷（通信作者），男，1973年生，讲师。研究方向为无人驾驶，AGV移动机器人。E-mail:Jiezhang@hust.edu.cn。
• 作者简介:林鑫，男，1996年生，硕士研究生。研究方向为视觉SLAM，AGV移动机器人。E-mail:linxin@hust.edu.cn。

Highly Robust Two-stage LiDAR-IMU External Parameter Online Calibration Algorithm

LIN Xin1;ZHANG Jie1;FENG Jingyi2;MENG Jie3;WANG Shuting3

1. 1.School of Energy and Power Engineering,Huazhong University of Science and Technology,Wuhan,430074
2.Basic Platform Research and Development,UISEE Technology Co.,Ltd.,Beijing,102400
3.School of Mechanical Science and Engineering,Huazhong University of Science and Technology,Wuhan,430074
• Online:2022-12-25 Published:2023-01-10

Abstract: Aiming at the problems that the degraded sensor data were difficult to calibrate external parameters when the vehicles were moving in an approximate plane, a highly robust two-stage LiDAR-IMU external parameters online calibration algorithm was proposed. The calibration algorithm included two stages: initial value calculation and online iterative optimization using nonlinear sliding window. In the first stage, the outliers in the pre-data set were eliminated, the hand-eye calibration model that contained only the rotation components was solved multiple times in the form of a sliding window, meanwhile, the conditions of the analytical solution screening were improved, the singular value decomposition（SVD） analytical solution of multiple weighted rotation external parameters was solved. The second stage minimized the residual function containing the external parameters, and used the rotational analytical solution as the initial value sliding window to iteratively optimize the six-degree-of-freedom external parameters, so that the external parameters converged quickly. In order to avoid the degradation of external parameters, when the degenerate motion and error history constraints were too large, the external parameters would be fixed. Compared with the original algorithm, the algorithm is robust to degraded sensor data, which may achieve accurate and robust online calibration of external parameters without the initial values of the external parameters.