Generalized Fringe-to-Phase Framework for Single-Shot 3D Reconstruction Integrating Structured Light with Deep...

Area of Science:

• Computer Vision
• 3D Reconstruction
• Artificial Intelligence

Background:

• Three-dimensional (3D) shape acquisition from single images is crucial for fields like medical imaging, robotics, and virtual reality.
• Existing methods often require multiple images or complex setups, limiting their application scope.
• Robust and efficient single-shot 3D reconstruction remains a significant challenge.

Purpose of the Study:

• To develop a robust 3D shape reconstruction approach from a single structured-light image.
• To integrate structured-light techniques with deep learning for enhanced accuracy and efficiency.
• To provide a versatile tool for various scientific and engineering applications.

Main Methods:

• A novel single-input dual-output deep neural network was designed.
• The network transforms a single structured-light image into intermediate fringe patterns and a coarse phase map.
• Ground-truth data was generated using a conventional fringe projection technique for training.

Main Results:

• The proposed approach accurately determines unwrapped phase distributions containing depth information.
• The method demonstrated robustness in experimental assessments.
• The system successfully reconstructs 3D shapes from single-shot structured-light images.

Conclusions:

• The integrated structured-light and deep learning approach offers a robust solution for single-shot 3D shape acquisition.
• This technique significantly advances the capabilities for 3D reconstruction in demanding applications.
• The method presents a promising tool for scientific research and engineering.