Passive binary defocusing for large depth 3D measurement based on deep learning.

Area of Science:

• Optics and Photonics
• Computer Vision
• Artificial Intelligence

Background:

• Phase-shifting profilometry (PSP) is a high-speed 3D measurement technique.
• Traditional PSP methods rely on manual projector defocusing, which is inaccurate and inflexible.
• The accuracy of traditional methods is often limited by fringe pitch and projector settings.

Purpose of the Study:

• To propose a novel deep learning-based method for generating high-quality sinusoidal patterns for PSP.
• To overcome the limitations of manual defocusing adjustments in traditional PSP.
• To enhance the accuracy, robustness, and flexibility of PSP measurements.

Main Methods:

• A convolutional neural network (CNN) was designed to learn and generate appropriate binary patterns.
• The method utilizes a passive defocusing strategy based on deep learning, eliminating manual adjustments.
• High-quality three-step sinusoidal patterns were generated using the trained CNN.

Main Results:

• The proposed deep learning method reduced phase error by 80%-90% across various fringe pitches.
• The technique demonstrated superior accuracy and robustness compared to traditional manual defocusing methods.
• Effective 3D measurements were achieved within a large measuring depth without projector defocus.

Conclusions:

• Deep learning offers a more accurate and flexible alternative for generating phase-shifting patterns in PSP.
• The proposed passive defocusing method significantly enhances measurement quality and reduces errors.
• This AI-driven approach advances the capabilities of high-speed 3D optical metrology.