Comparative assessment of two algorithms for calibrating stereophotogrammetric systems

Area of Science:

• Photogrammetry
• Computer Vision
• Metrology

Background:

• Accurate camera calibration is crucial for 3D reconstruction.
• Existing algorithms like Direct Linear Transform (DLT) have limitations, especially with lens distortion and extrapolated data.
• Novel algorithms are needed to improve robustness and accuracy.

Purpose of the Study:

• Compare the performance of two photogrammetric algorithms: Marzan and Karara's DLT and the author's CESNO (close to Modified DLT).
• Evaluate their capability for camera calibration and 3D point reconstruction, particularly when extrapolating beyond control point distributions.
• Assess the impact of lens distortion and varying camera parameters on algorithm accuracy.

Main Methods:

• Computer simulation of a two-camera stereophotogrammetric system.
• Testing with various internal camera parameters, including scaled principal distances.
• Simulating low, medium, and strong non-linear lens distortion.
• Investigating the effect of reducing the number of control points.

Main Results:

• CESNO demonstrates superior accuracy compared to DLT, especially under medium to strong lens distortion.
• CESNO's advantage is pronounced when the two cameras have significantly different principal distances.
• Algorithm performance degrades with fewer control points, but CESNO remains more robust.

Conclusions:

• CESNO is a more robust and accurate algorithm for camera calibration and 3D reconstruction than DLT.
• The algorithm shows particular strength in handling non-linear lens distortion and differing camera configurations.
• CESNO offers improved performance for photogrammetric applications facing challenging imaging conditions.