[Paper Review] A Novel Dual-Lidar Calibration Algorithm Using Planar Surfaces

Jianhao Jiao, Qinghai Liao+6 more|arXiv (Cornell University)|Apr 27, 2019

Robotics and Sensor-Based Localization24 references4 citations

TL;DR

This paper proposes a novel dual-lidar calibration algorithm that uses three linearly independent planar surfaces in the environment as natural calibration targets, eliminating the need for artificial markers or additional sensors. It employs a closed-form solver for initialization and nonlinear optimization to refine extrinsic parameters, achieving sub-0.05 rad rotation and sub-0.1 m translation errors in both simulation and real-world experiments.

ABSTRACT

Multiple lidars are prevalently used on mobile vehicles for rendering a broad view to enhance the performance of localization and perception systems. However, precise calibration of multiple lidars is challenging since the feature correspondences in scan points cannot always provide enough constraints. To address this problem, the existing methods require fixed calibration targets in scenes or rely exclusively on additional sensors. In this paper, we present a novel method that enables automatic lidar calibration without these restrictions. Three linearly independent planar surfaces appearing in surroundings is utilized to find correspondences. Two components are developed to ensure the extrinsic parameters to be found: a closed-form solver for initialization and an optimizer for refinement by minimizing a nonlinear cost function. Simulation and experimental results demonstrate the high accuracy of our calibration approach with the rotation and translation errors smaller than 0.05rad and 0.1m respectively.

Motivation & Objective

• To enable automatic, marker-free calibration of dual lidars in outdoor environments.
• To overcome the limitations of existing methods that rely on artificial calibration targets or additional sensors.
• To provide a robust and accurate calibration solution using only planar features from the environment.
• To ensure high accuracy with minimal human intervention through a two-stage optimization process.
• To validate the method under both ideal and challenging real-world scenarios with occlusions and clutter.

Proposed method

• Uses RANSAC-based plane extraction to detect three linearly independent planar surfaces from lidar scans.
• Establishes geometric constraints between the three planes to derive a closed-form solution for initial extrinsic parameter estimation.
• Applies a nonlinear optimization framework to refine the initial estimate by minimizing a cost function based on plane-to-plane alignment.
• Relies on the structural consistency of planar surfaces forming a corner-like configuration to ensure stable correspondences.
• Uses a single-shot measurement for calibration, enabling real-time applicability.
• Implements a robust initialization that avoids dependence on motion data or user-provided initial guesses.

Experimental results

Research questions

• RQ1 Can dual-lidar calibration be achieved without artificial calibration targets or additional sensors?
• RQ2 How accurately can extrinsic parameters be estimated using only natural planar surfaces in the environment?
• RQ3 Can the method maintain high accuracy under challenging conditions such as occlusions and clutter?
• RQ4 How does the proposed two-stage approach (closed-form + optimization) compare to ICP or motion-based methods in terms of accuracy and robustness?
• RQ5 What is the computational cost and convergence behavior of the proposed optimization in real-world scenarios?

Key findings

• The proposed method achieves rotation and translation errors below 0.05 rad and 0.1 m, respectively, in both simulated and real-world experiments.
• For the $L_1 \ominus L_2$ configuration, the method achieved a rotation error of 0.0203 rad and translation error of 0.067196 m.
• For the $L_1 \ominus L_3$ configuration, the method achieved a rotation error of 0.0238 rad and translation error of 0.08357 m.
• In a hard scenario with occlusions, the method achieved a rotation error of 0.0292 rad and translation error of 0.0930 m, demonstrating robustness.
• The optimization phase took approximately 27.3s and 35.8s for the two configurations, indicating acceptable computational cost.
• Qualitative results show well-aligned point clouds with minimal uncertainty after calibration, confirming accurate registration.