Estimation of extrinsic parameters for dynamic binocular stereo vision using unknown-sized rectangle images.

Area of Science:

• Computer Vision
• Robotics
• 3D Reconstruction

Background:

• Accurate extrinsic parameter calibration is crucial for dynamic binocular stereo vision systems.
• Existing methods often require controlled environments or specific calibration objects.
• Real-time calibration is essential for applications involving moving cameras.

Purpose of the Study:

• To develop an online calibration method for extrinsic parameters of dynamic binocular stereo vision.
• To enable calibration for cameras with unknown-sized rectangular targets and free rotation (two-axis).
• To achieve precise estimation of camera yaw and pitch angles.

Main Methods:

• Utilizes an Ellipse-A attitude and heading reference system for initial angle approximation.
• Employs interimage homography with a single rectangular centroid for iterative angle solving.
• Refines pitch and yaw angles using the fundamental matrix property with corresponding rectangular centroids.

Main Results:

• Achieved root mean square (rms) error for attitude angles below 0.12°.
• Achieved rms error for translation vector below 7.2 mm.
• Demonstrated rms relative error for 3D coordinates not exceeding 1.6%.

Conclusions:

• The proposed online calibration method is efficient and accurate for dynamic binocular stereo vision.
• The method effectively calibrates cameras with free rotation and unknown-sized rectangular targets.
• Validated through simulations and real experiments, outperforming existing methods like SIFT and Zhang's.