Error analysis of pure rotation-based self-calibration

Area of Science:

• Computer Vision
• Robotics
• Photogrammetry

Background:

• Self-calibration is a standard method for determining intrinsic camera parameters.
• Pure rotation is assumed for reliable self-calibration, but is difficult to achieve in practice.
• Unwanted translation during camera motion introduces errors in intrinsic parameter recovery.

Purpose of the Study:

• To analyze the impact of translation on intrinsic camera parameter accuracy during self-calibration.
• To derive error expressions for recovered parameters under translational motion.
• To identify conditions under which translation-independent solutions are feasible.

Main Methods:

• Derived closed-form error expressions for intrinsic parameters from a single image pair with non-degenerate motion.
• Conducted repeated experiments to analyze multi-rotation scenarios lacking closed-form solutions.
• Evaluated the influence of translational motion on camera self-calibration accuracy.

Main Results:

• Quantified errors in intrinsic camera parameters caused by translational motion.
• Demonstrated the existence of translation-independent solutions under specific practical conditions.
• Provided insights into selecting optimal self-calibration strategies based on motion characteristics.

Conclusions:

• The presence of translation significantly affects the accuracy of self-calibrated intrinsic camera parameters.
• Translation-independent solutions can be achieved, offering robustness in real-world applications.
• The error analysis aids in choosing the most reliable self-calibration method for given scenarios.