A Novel Calibration Board and Experiments for 3D LiDAR and Camera Calibration

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Traditional 3D LiDAR and camera calibration methods suffer from diverse calibration board designs, incomplete feature extraction, and significant errors.
• Accurate extrinsic and intrinsic parameter calibration is crucial for reliable sensor fusion in applications like autonomous driving and robotics.

Purpose of the Study:

• To address the limitations of existing calibration techniques by introducing a novel calibration board and an improved feature point extraction method.
• To achieve highly accurate and stable calibration of 3D Light Detection and Ranging (LiDAR) and camera systems.

Main Methods:

• Design of a novel calibration board (BWDC) incorporating local gradient depth and main plane corner information.
• Development of a "three-step fitting interpolation method" for precise feature point selection and coordinate extraction in both LiDAR and camera systems.
• Experimental validation using incremental verification and reprojection error analysis.

Main Results:

• The proposed BWDC and fitting method achieved accurate coordinate transformation matrices and sensor parameters, with dynamic changes within 0.2% in repeatable experiments.
• Incremental verification showed a small difference (approx. 0.5%) between experimental and actual values.
• The average reprojection error was 1.8312 pixels, with minimal variation (≤0.1 pixels) across different distances, indicating high accuracy and stability.

Conclusions:

• The novel BWDC and "three-step fitting interpolation method" offer a significant improvement in 3D LiDAR and camera calibration accuracy and stability.
• The method provides a reliable solution for obtaining precise sensor parameters, essential for advanced perception systems.
• The demonstrated accuracy and stability make this approach suitable for dynamic and demanding real-world applications.