Per-Pixel Calibration Based on Multi-View 3D Reconstruction Errors Beyond the Depth of Field

Area of Science:

• Microscopic Imaging
• Optical Metrology
• 3D Reconstruction

Background:

• Shallow depth of field in 3D microscopy challenges accurate 3D reconstruction, especially with defocus.
• Traditional calibration methods using feature points degrade under defocus, limiting performance.
• Existing techniques struggle to extend calibration volume while maintaining accuracy in defocused conditions.

Purpose of the Study:

• To develop a per-pixel calibration method for 3D microscopic imaging that overcomes defocus challenges.
• To enhance 3D reconstruction accuracy and extend calibration volume in defocused scenarios.
• To introduce a novel calibration approach based on multi-view 3D reconstruction errors.

Main Methods:

• Proposed a per-pixel calibration method utilizing 3D reconstruction errors from multiple binocular setups.
• Analyzed multi-view 3D reconstruction error distributions using a microscopic 3D measurement system with telecentric lenses.
• Developed a 3D proportion model for error-based per-pixel calibration, derived from error distributions.

Main Results:

• The proposed method demonstrated robust convergence across multiple binocular setups.
• Near focus, multi-view 3D reconstruction error was <0.5 camera pixel pitch, with accuracy within 0.5% of the measurement range.
• Beyond tenfold depth of field, error increased to <2 camera pixel pitches, with accuracy within 1% of the measurement range.

Conclusions:

• The novel per-pixel calibration method effectively addresses defocus challenges in 3D microscopic imaging.
• The approach achieves high-precision 3D reconstruction, maintaining accuracy over an extended depth of field.
• Experimental results validate the feasibility and accuracy of the proposed error-based calibration technique.