Multi-Camera Calibration Using Far-Range Dual-LED Wand and Near-Range Chessboard Fused in Bundle Adjustment

Area of Science:

• Computer Vision
• Metrology
• Robotics

Background:

• Accurate 3D reconstruction requires precise calibration of multiple synchronized cameras.
• Existing calibration methods, like wand-based approaches, can suffer from overfitting and reduced accuracy.

Purpose of the Study:

• To develop and validate a robust calibration approach for multi-camera systems in large capture volumes.
• To address and mitigate the overfitting issue in wand-based camera calibration.
• To improve the overall 3D accuracy and consistency of camera calibration.

Main Methods:

• Utilized an active wand with LED markers for data capture within the volume.
• Combined wand data with close-range chessboard images for intrinsic and extrinsic calibration.
• Employed a novel bundle adjustment method incorporating chessboard constraints.
• Benchmarked 3D triangulation accuracy against an independent ground-truth trajectory.

Main Results:

• The proposed method significantly improved 3D accuracy and result consistency compared to traditional wand-based calibration.
• Identified and addressed an overfitting issue inherent in wand-only calibration techniques.
• Demonstrated the superiority of ray distance-based (RDB) triangulation over direct linear transformation (DLT).

Conclusions:

• The integrated calibration approach enhances the reliability of 3D applications using synchronized multi-camera systems.
• A new benchmarking workflow and dataset are provided for future research and method comparison.
• The findings offer practical improvements for 3D measurement accuracy and consistency.