Deep learning based measurement accuracy improvement of high dynamic range objects in fringe projection profilometry

Area of Science:

• Optics and Photonics
• Computer Vision
• Artificial Intelligence

Background:

• Accurate 3D measurement using fringe projection profilometry (FPP) is crucial.
• High dynamic range (HDR) objects present challenges in FPP due to fringe saturation and low contrast.
• Traditional FPP methods require numerous images for HDR objects, limiting their use for dynamic scenes.

Purpose of the Study:

• To develop a novel deep learning-based method for accurate phase demodulation in FPP for HDR objects.
• To overcome limitations of traditional methods in measuring objects with significant variations in reflectivity.
• To enable faster and more reliable 3D measurements of moving HDR objects.

Main Methods:

• An improved UNet deep neural network was employed to establish a 'many-to-one' mapping for phase retrieval.
• Utilized π-shifted binary fringes to capture enhanced fringe information, particularly in saturated regions.
• Implemented a deep learning approach for efficient phase demodulation.

Main Results:

• The proposed method successfully retrieves the wrapped phase of HDR objects with high accuracy.
• Demonstrated effectiveness and reliability in experimental validation.
• Achieved faster phase retrieval compared to traditional FPP techniques for HDR objects.

Conclusions:

• The deep learning-based method offers a significant advancement for 3D measurement of HDR objects using FPP.
• The approach effectively addresses challenges of fringe saturation and low contrast.
• This technique provides a robust solution for accurate and rapid 3D reconstruction of challenging surfaces.