Multi-view fringe projection profilometry based on phase texture and U-Net.

Area of Science:

• Optics and Photonics
• Computer Vision
• Metrology

Background:

• Multi-view fringe projection profilometry (FPP) relies on pattern separability for 3D reconstruction.
• Traditional FPP methods demand numerous patterns, limiting projection efficiency.

Purpose of the Study:

• To develop a novel phase texture technique for multi-view FPP to reduce pattern count.
• To enhance projection efficiency and improve 3D reconstruction accuracy in FPP.

Main Methods:

• Encoding wrapped phase as an aperiodic texture to reduce pattern numbers.
• Utilizing a U-Net neural network trained on virtual datasets with noise and defocus as a decoder.
• Designing a monotonic loss function to improve decoder adaptability for real-world measurements.

Main Results:

• The proposed phase texture patterns are separable, enabling efficient 3D reconstruction.
• Reconstructions were achieved using only one-fifth the number of patterns compared to traditional methods.
• U-Net demonstrated improved decoding performance with the monotonic loss function and enhanced dataset.

Conclusions:

• The phase texture technique significantly enhances projection efficiency in multi-view FPP.
• The U-Net decoder, trained with a monotonic loss function, shows robust performance for real measurements.
• This approach offers a more efficient and accurate method for 3D shape measurement using FPP.