Deep-learning-enabled single-shot fringe projection profilometry based on inner shifting-phase encoding.

Area of Science:

• Optical Metrology
• Computer Vision
• Artificial Intelligence

Background:

• Single-shot fringe projection profilometry (FPP) is crucial for dynamic 3D measurements.
• Current deep learning methods struggle with simultaneous wrapped phase and fringe order extraction.
• Ambiguity arises from estimating continuous phase and discrete fringe order from a single image, especially for isolated objects.

Purpose of the Study:

• To develop a deep learning-enabled single-shot FPP method for accurate absolute phase recovery.
• To address the inherent ambiguity in joint phase and fringe order estimation.
• To enable robust 3D measurement of complex and isolated objects in dynamic scenarios.

Main Methods:

• Proposed a physically constrained dual-network architecture integrating inner shifting-phase encoding.
• Embedded fringe order into the phase-shifting component of a single fringe pattern.
• Employed two lightweight convolutional neural networks for decoupled phase and fringe order regression and classification.

Main Results:

• Achieved high-accuracy absolute phase recovery comparable to multi-frame techniques.
• Demonstrated unambiguous phase recovery for complex and isolated objects.
• Maintained low computational complexity and showed potential for dynamic 3D measurement.

Conclusions:

• The proposed physically consistent decoupled design overcomes optimization conflicts in end-to-end regression.
• The method offers a robust solution for single-shot FPP, improving accuracy and efficiency.
• Validated potential for real-time dynamic 3D measurement applications.