Absolute phase retrieval of shiny objects using fringe projection and deep learning with computer-graphics-based...

Area of Science:

• Optics and Photonics
• Computer Vision
• Artificial Intelligence

Background:

• Fringe projection profilometry is a key 3D measurement technique.
• Nonlinear fringe distortion on shiny surfaces causes significant errors.
• Existing methods struggle with specular reflection and complex geometries.

Purpose of the Study:

• To develop a high-precision 3D measurement method for shiny objects.
• To overcome the limitations of traditional fringe projection profilometry on metallic and glossy surfaces.
• To reduce measurement errors caused by nonlinear fringe distortions.

Main Methods:

• Utilized computer graphics (CG) to simulate fringe projection in a controlled virtual environment.
• Trained a deep neural network (DNN) using CG-generated fringe data from shiny objects.
• Applied the trained DNN to correct nonlinear distortions in real-world measurements.

Main Results:

• The proposed method effectively reduces nonlinear fringe distortion caused by gloss.
• Achieved significantly higher measurement accuracy for shiny objects compared to conventional methods.
• Demonstrated the capability of deep learning to handle complex optical phenomena in 3D metrology.

Conclusions:

• Computer graphics and deep learning offer a robust solution for 3D measurement of challenging surfaces.
• The proposed technique enhances the applicability of fringe projection profilometry to industrial inspection of metallic components.
• This approach paves the way for more accurate and reliable 3D surface reconstruction in various fields.