3D LiDAR Extrinsic Calibration Using Plane Matching Across Viewpoints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extrinsic parameter calibration of multiple 3D LiDAR sensors in autonomous navigation systems face challenges, particularly in accurately determining relative positions without environmental reorganization and are sensitive to sensor uncertainty and varying viewpoints, which complicates calibration and reduces robustness.
Innovation Solution
An extrinsic calibration method that extracts target planes from point clouds using plane extraction algorithms like region growing segmentation and MLESAC, calculates initial parameters based on plane similarity, and optimizes final parameters to minimize measurement variance using a Levenberg-Marquardt technique, allowing for calibration without artificial marks and ensuring robustness across different viewpoints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If multiple low-resolution 3D LiDAR sensors are used to reduce system cost and measure wide environments, then system cost is reduced and coverage area is increased, but measurement precision and calibration accuracy deteriorate due to narrow viewing angles and low point cloud density
Solution Approach 1:
The calibration process is segmented into two distinct phases: coarse calibration using plane extraction algorithms to establish initial relative positions, and fine calibration using ICP algorithm to refine the calibration accuracy. This segmentation allows the system to overcome the low precision of individual low-resolution sensors by combining multiple calibration approaches.
Solution Approach 2:
Plane extraction algorithms serve as an intermediary step between raw point cloud data and final calibration results. By extracting plane information as an intermediate representation, the system can achieve more accurate calibration than direct point-to-point matching, bridging the gap between low-resolution sensor data and high-precision calibration requirements.
2Ease of manufacture
If extrinsic parameter calibration is performed using feature information from the surrounding environment, then calibration can be done without artificial marks, but reliability deteriorates due to sensitivity to sensor uncertainty and varying viewpoints
Solution Approach 1:
The system performs preliminary plane extraction and correspondence identification before executing the ICP algorithm. By pre-processing the point cloud data to identify and match plane features between sensors, the system establishes a reliable initial framework that makes the subsequent calibration process more robust to sensor uncertainties and viewpoint variations.
Solution Approach 2:
The calibration process uses feedback from plane extraction results to guide the ICP algorithm. The initial calibration parameters obtained from plane matching are fed back as starting points for the iterative ICP optimization, allowing the system to continuously refine calibration accuracy while maintaining robustness against environmental variations.
3Ease of operation
If plane extraction algorithms are used for initial calibration, then calibration can be performed without environmental reorganization, but device complexity increases due to the need for multiple algorithmic steps
Solution Approach 1:
The calibration system is self-sufficient, using only the point cloud data captured by the LiDAR sensors themselves without requiring external calibration objects or environmental modifications. The plane extraction algorithms automatically identify calibration features from the environment that the sensors naturally observe, making the system self-calibrating and operationally simple despite the computational complexity behind the scenes.
Data Source
AI summary
An extrinsic calibration method of multiple 3D LiDAR sensors for an autonomous navigation system is proposed, the method including collecting point clouds by each of the multiple 3D LiDAR sensors; extracting multiple target planes corresponding to a plane from the point clouds of each of the 3D LiDAR sensors; by using the target plane of any one of the multiple 3D LiDAR sensors as a reference plane, detecting a corresponding plane from the target planes of each of the remaining of the 3D LiDAR sensors on the basis of a similarity with the reference plane; calculating initial extrinsic parameters for matching between the reference plane and the corresponding plane based on plane parameters of the reference plane and corresponding plane corresponding to each other; and calculating final extrinsic parameters that minimize variance of measurement points for the reference plane and corresponding plane, on the basis of the initial extrinsic parameters.