Camera-LIDAR Calibration Using Checkerboard Corner Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current camera-LIDAR calibration methods, such as the KITTI method, are inadequate for autonomous vehicles with sparse LIDAR data or when objects are far from the LIDAR unit, as they rely on dense LIDAR beams and fail to accurately calibrate due to environmental noise and complexity in matching LIDAR and image surfaces.
Innovation Solution
A method that involves filtering 3D point clouds from a LIDAR unit and 2D images from a camera to estimate corner positions of a checkerboard in both LIDAR and image coordinate systems, using Principal Component Analysis (PCA) to eliminate depth and interpolate inner corner positions, generating a transformation matrix for accurate calibration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the KITTI method is used for camera-LIDAR calibration, then calibration can be performed with dense LIDAR data, but the method fails when LIDAR data is sparse or objects are far from the LIDAR unit
Solution Approach 1:
The patent changes the fundamental parameters of the calibration approach by transitioning from surface norm vector-based methods to corner point-based methods. This parameter change enables the system to work effectively with sparse LIDAR data by focusing on discrete, easily detectable corner points rather than requiring dense surface coverage for normal vector calculation
Solution Approach 2:
The patent extracts only the essential corner points of the checkerboard from the LIDAR point cloud, eliminating the need to process entire surfaces. This extraction approach reduces computational complexity and enables reliable calibration even when LIDAR data is sparse, as only the corner points need to be detected and matched
2Measurement precision
If surface norm vectors are created based on nearest points, then calibration may work with dense LIDAR data, but the method fails when distance between LIDAR beams is large
Solution Approach 1:
The patent extracts only corner points from the LIDAR data, eliminating the need to compute surface norm vectors from multiple nearest points. This extraction simplifies the computational process and maintains precision by focusing on the distinctive corner features of the checkerboard rather than requiring dense surface sampling
Solution Approach 2:
The patent uses image processing to accurately detect corner positions in the 2D image, then copies these detected corner positions to guide the search in LIDAR space. This copying approach ensures precise corner identification without requiring complex 3D surface analysis
3Measurement precision
If multiple checkerboards are placed at different positions and angles, then calibration accuracy can be improved, but the complexity of the calibration process increases
Solution Approach 1:
The patent creates a universal calibration pipeline that handles checkerboards at any position and angle through a standardized corner detection and matching process. The method universally applies to single or multiple checkerboards, eliminating the need for different processing approaches for different configurations and reducing overall system complexity
Data Source
AI summary
A system and method for performing camera-LIDAR calibration based on a checkerboard placed in proximity to a vehicle, the method includes: receiving a 3D point cloud and a 2D image including the checkerboard; filtering the 3D point cloud representing the checkerboard; converting the filtered 3D point cloud to a 2D point cloud in a translated coordinate system; estimating a 2D position, in the translated coordinate system, for each outer corner of the checkerboard represented by the 2D point cloud; estimating a 2D position in the translated coordinate system for each inner corner of the checkerboard represented by the 2D point cloud; determining a 3D position, in a LIDAR coordinate system, for each corner of the checkerboard in the 3D point cloud based on the corresponding 2D position in the translated coordinate system; and determining a 2D position of each corner of the checkerboard in a 2D image coordinate system.