Single-Model Self-Recovering Fringe Projection Profilometry Absolute Phase Recovery Method Based on Deep Learning

Area of Science:

• Optical Metrology
• Computer Vision
• Artificial Intelligence

Background:

• Fringe Projection Profilometry (FPP) faces challenges in achieving high-resolution absolute phase recovery with single-shot measurements.
• Existing methods often require multiple measurements or complex processing for accurate phase retrieval.

Purpose of the Study:

• To develop a single-model, deep learning-based method for efficient and accurate absolute phase recovery in FPP.
• To enable high-resolution 3D shape reconstruction from a single fringe image under complex lighting conditions.

Main Methods:

• A novel Fringe Prediction Self-Recovering network was developed to convert a single fringe image into multiple self-recovering fringe images.
• A self-recovering algorithm was employed for wrapped phase and fringe grade determination.
• A virtual measurement system was utilized for efficient dataset preparation.

Main Results:

• The method achieved high-resolution absolute phase recovery with a Mean Absolute Error (MAE) of 0.015 rad in a physical system.
• Reconstructed point clouds exhibited a Root Mean Square Error (RMSE) of 0.02 mm.
• The deep learning model demonstrated robustness and generalization across various lighting conditions and measurement scenarios.

Conclusions:

• The proposed deep learning approach simplifies phase retrieval and unwrapping for FPP.
• It offers a faster, more efficient, and lightweight solution for online 3D detection compared to existing methods.
• The technique is effective even under complex ambient lighting, providing a valuable reference for FPP applications.