Composite fringe projection deep learning profilometry for single-shot absolute 3D shape measurement.

Area of Science:

• Optics and Photonics
• Computer Vision
• Metrology

Background:

• Single-shot fringe projection profilometry (FPP) is crucial for capturing dynamic 3D scenes.
• Achieving high-precision, absolute 3D reconstruction with a single pattern remains a significant challenge.
• Frequency-multiplexing (FM) offers a path to single-shot absolute measurements but suffers from spectrum aliasing.

Purpose of the Study:

• To develop a novel single-shot FPP method overcoming spectrum aliasing for precise 3D reconstruction.
• To introduce Composite Fringe Projection Deep Learning Profilometry (CDLP) by integrating deep learning with FM techniques.
• To validate the CDLP method's effectiveness in both static and dynamic 3D measurement scenarios.

Main Methods:

• Application of deep learning, specifically an improved deep convolutional neural network, to frequency multiplexing composite fringe projection.
• Combining physical modeling with data-driven approaches for robust phase retrieval.
• Training the neural network to directly process single-shot composite fringe images for phase unwrapping.

Main Results:

• Demonstrated high-precision and unambiguous phase retrieval from single composite fringe images.
• Successfully avoided spectrum aliasing issues inherent in traditional FM methods.
• Achieved high-quality absolute 3D surface reconstruction in both static and dynamic environments.

Conclusions:

• CDLP offers a robust solution for single-shot absolute 3D measurement in dynamic scenes.
• The deep learning approach effectively addresses spectrum aliasing, enabling precise phase retrieval.
• This method reconstructs high-quality 3D surfaces using only one composite fringe image.