Single-shot absolute 3D shape measurement with deep-learning-based color fringe projection profilometry

Area of Science:

• Optics and Photonics
• Computer Vision
• Metrology

Background:

• Fringe Projection Profilometry (FPP) aims for high-resolution 3D surface recovery from single images.
• Color FPP offers potential but faces challenges like color imbalance and crosstalk.
• Deep learning has shown promise in advancing FPP techniques.

Purpose of the Study:

• To develop a single-shot absolute 3D shape measurement method using deep learning-based color FPP.
• To overcome limitations of traditional color FPP, including color imbalance and crosstalk.
• To achieve high-accuracy phase retrieval and robust phase unwrapping from a single image.

Main Methods:

• A deep neural network trained on extensive datasets.
• Utilizing the three-channel multiplexing properties of color fringe projection.
• Implementing a single-shot data acquisition strategy.
• Applying deep learning for direct prediction of absolute phase from a single color fringe image.

Main Results:

• High-resolution, motion-artifact-free, and crosstalk-free absolute phase prediction.
• More accurate phase retrieval compared to traditional methods.
• More robust phase unwrapping capabilities.
• Demonstrated high-accuracy single-frame absolute 3D shape measurement for complex objects.

Conclusions:

• Deep learning-based color FPP enables accurate single-shot absolute 3D shape measurement.
• The proposed method effectively addresses limitations of traditional FPP techniques.
• This approach offers a significant advancement for 3D metrology applications.