Continuous stereo self-calibration by camera parameter tracking.

Area of Science:

• Computer Vision
• Robotics
• Photogrammetry

Background:

• Stereo reconstruction accuracy is sensitive to camera calibration uncertainties.
• Continuous self-calibration is crucial for dynamic environments and long-term operation.
• Existing methods may struggle with moving objects or require extensive computation.

Purpose of the Study:

• To develop a consistent framework for continuous stereo self-calibration.
• To identify and track critical calibration parameters.
• To improve the accuracy and robustness of stereo camera calibration.

Main Methods:

• Utilized bundle adjustment with reduced structure representation for accurate parameter estimation.
• Employed epipolar and trilinear constraints for geometric validation.
• Implemented an iterated extended Kalman filter for recursive calibration refinement.
• Formulated optimization criteria using a Gauss-Helmert model for high accuracy.

Main Results:

• Demonstrated that reduced order bundle adjustment provides highly accurate results.
• Showcased the epipolar constraint's ability to yield instantaneous calibration, unaffected by moving objects.
• Validated the complementary nature of different geometric constraints.
• Proposed a combined approach leveraging the strengths of multiple constraints.

Conclusions:

• A combined stereo self-calibration framework offers superior accuracy and robustness.
• The epipolar constraint stabilizes and accelerates the calibration process.
• The proposed method is effective for active stereo systems and diverse imagery.