Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Rotation Initialization and Stepwise Refinement for Universal LiDAR Calibration

Yifan Duan, Xinran Zhang, Guoliang You, Yilong Wu, Xingchen Li, Yao Li, Xiaomeng Chu, Jie Peng, Yu Zhang, Jianmin Ji, Yanyong Zhang

Abstract

Autonomous systems often employ multiple LiDARs to leverage the integrated advantages, enhancing perception and robustness. The most critical prerequisite under this setting is the estimating the extrinsic between each LiDAR, i.e., calibration. Despite the exciting progress in multi-LiDAR calibration efforts, a universal, sensor-agnostic calibration method remains elusive. According to the coarse-to-fine framework, we first design a spherical descriptor TERRA for 3-DoF rotation initialization with no prior knowledge. To further optimize, we present JEEP for the joint estimation of extrinsic and pose, integrating geometric and motion information to overcome factors affecting the point cloud registration. Finally, the LiDAR poses optimized by the hierarchical optimization module are input to time synchronization module to produce the ultimate calibration results, including the time offset. To verify the effectiveness, we conduct extensive experiments on eight datasets, where 16 diverse types of LiDARs in total and dozens of calibration tasks are tested. In the challenging tasks, the calibration errors can still be controlled within 5cm and 1° with a high success rate.

Rotation Initialization and Stepwise Refinement for Universal LiDAR Calibration

Abstract

Autonomous systems often employ multiple LiDARs to leverage the integrated advantages, enhancing perception and robustness. The most critical prerequisite under this setting is the estimating the extrinsic between each LiDAR, i.e., calibration. Despite the exciting progress in multi-LiDAR calibration efforts, a universal, sensor-agnostic calibration method remains elusive. According to the coarse-to-fine framework, we first design a spherical descriptor TERRA for 3-DoF rotation initialization with no prior knowledge. To further optimize, we present JEEP for the joint estimation of extrinsic and pose, integrating geometric and motion information to overcome factors affecting the point cloud registration. Finally, the LiDAR poses optimized by the hierarchical optimization module are input to time synchronization module to produce the ultimate calibration results, including the time offset. To verify the effectiveness, we conduct extensive experiments on eight datasets, where 16 diverse types of LiDARs in total and dozens of calibration tasks are tested. In the challenging tasks, the calibration errors can still be controlled within 5cm and 1° with a high success rate.
Paper Structure (39 sections, 20 equations, 19 figures, 7 tables, 1 algorithm)

This paper contains 39 sections, 20 equations, 19 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. LiDAR-LiDAR Calibration
  4. Geometry-based methods
  5. Motion-based methods
  6. Point Cloud Descriptor
  7. BEV Projection Based Methods
  8. Spherical Projection Based Methods
  9. Multi-LiDAR SLAM
  10. Preliminaries
  11. Point Cloud Registration
  12. SLAM
  13. Hand-Eye Method
  14. System Overview
  15. Rotation Initialization
  16. ...and 24 more sections

Figures (19)

  • Figure 1: The illustration of seven types of LiDAR in the simulated dataset. Mechanical LiDARs exhibit varying densities, solid-state LiDARs have different FOV, and mechanical and solid-state LiDARs have distinctly different point cloud distributions. This diversity presents new challenges for LiDAR calibration. This paper aims to introduce a universal, sensor-agnostic calibration method.
  • Figure 2: The pipeline of the universal LiDAR calibration involving four core steps, i.e., TERRA based rotation initialization, JEEP for joint estimation of extrinsic and poses, hierarchical optimization based pose refinement and time synchronization for enhancing the accuracy of extrinsic parameters and estimating the time offset.
  • Figure 3: Instructions for the generation of scan context. By projecting the point cloud onto the $x-y$ plane and dividing it into bins based on azimuth and radius, a two-dimensional descriptor is generated.
  • Figure 4: Instructions for the generation of TERRA. The process is divided into three steps: sampling the sphere using the Fibonacci lattice, projecting the point cloud from LiDAR onto the unit sphere, and populating the values into TERRA based on the correspondences.
  • Figure 5: Instructions for rotation initialization. Each LiDAR frame has a TERRA extracted from it. Then the descriptor of one LiDAR is fixed, while rotating the other to find the optimal match. The rotation corresponding to the optimal match is the result of the rotation initialization. Different colors represent different values and the TERRA is a vector of length $N$.
  • ...and 14 more figures