Underwater structured-light 3D imaging method based on FP-DiffNet.

Area of Science:

• Optics and Photonics
• Computer Vision
• Robotics

Background:

• Underwater 3D imaging is crucial for scientific and engineering tasks.
• Degradation of fringe patterns due to light attenuation, scattering, and distortion limits accuracy.
• Existing methods struggle with accuracy in turbid or optically complex underwater environments.

Purpose of the Study:

• To develop a robust method for restoring degraded fringe patterns in underwater structured-light 3D imaging.
• To enhance the accuracy of phase reconstruction and 3D object reconstruction in challenging underwater conditions.
• To create a practical and adaptable solution for underwater 3D imaging without prior water optical property knowledge.

Main Methods:

• Analysis of structured light propagation and its impact on fringe image quality in water.
• Development of FP-DiffNet, a diffusion-model-based neural framework for fringe restoration.
• Training a U-Net model with physics-guided constraints and adaptive noise annealing for iterative denoising.

Main Results:

• FP-DiffNet effectively restores degraded fringe patterns, separating scattering-induced noise while preserving fringe structures.
• Achieved high performance in extremely turbid water: PSNR of 42.42 dB and wrapped-phase MAE of 0.0354 rad.
• Enabled sub-millimeter absolute measurement and 3D reconstruction with RMSE below 0.1 mm in turbid water.

Conclusions:

• The proposed FP-DiffNet framework offers a significant advancement in underwater structured-light 3D imaging.
• It provides robust and accurate 3D reconstruction capabilities in challenging, turbid underwater environments.
• The method's independence from water optical parameter modeling makes it highly practical for real-world applications.