Machine-Learning-Inspired Workflow for Camera Calibration.

Area of Science:

• Optical Metrology
• Computer Vision
• Geometric Measurement

Background:

• Modern digital cameras offer high performance for precise measurements.
• Camera calibration is critical for geometrical measurements but often lacks systematic methodology.
• Inconsistent data collection, model selection, and result interpretation hinder calibration accuracy.

Purpose of the Study:

• To introduce a systematic and metrologically sound approach for camera calibration.
• To characterize calibration outcomes with known and controlled quality.
• To improve the reliability of camera-assisted geometrical measurements.

Main Methods:

• Inspired by machine learning workflows and practical measurement needs.
• Utilized standard calibration tools combined with active targets and phase-shifted cosine patterns.
• Employed simple parametric models for characterizing imaging geometry.

Main Results:

• Achieved sub-millimeter uncertainty in characterizing industrial camera imaging geometry up to several meters.
• Demonstrated control over data quality and calibration parameters throughout the process.
• Validated the effectiveness of the systematic approach even with basic models.

Conclusions:

• A systematic, metrologically sound camera calibration method enhances measurement precision.
• The proposed approach ensures known and controllable data quality and parameter reliability.
• This methodology is crucial for advancing high-precision applications in optical metrology and computer vision.