High-precision dynamic three-dimensional shape measurement of specular surfaces based on deep learning

Area of Science:

• Optical Metrology
• Computer Vision
• Artificial Intelligence

Background:

• Traditional phase measuring deflectometry (PMD) faces challenges in balancing precision and speed for 3D surface reconstruction.
• Accurate and rapid measurement of specular surfaces is crucial for advanced manufacturing and quality control.

Purpose of the Study:

• To introduce an orthogonal encoding phase measuring deflectometry (PMD) method integrated with deep learning.
• To enable high-precision, high-speed 3D shape reconstruction of specular surfaces using single-frame distorted patterns.

Main Methods:

• Development of an orthogonal encoding technique for dynamic PMD.
• Application of deep learning algorithms to reconstruct 3D shapes from single-frame fringe patterns.
• Experimental validation comparing the proposed method against traditional multi-step phase-shifting techniques.

Main Results:

• The deep learning-based orthogonal encoding PMD achieves high-accuracy phase and shape information.
• Measurement accuracy closely approximates that of a ten-step phase-shifting method.
• The method demonstrates excellent performance in dynamic measurement scenarios.

Conclusions:

• Deep learning can be effectively combined with dynamic PMD for precise 3D reconstruction of specular surfaces.
• The proposed method offers a significant advancement in speed and precision for optical measurement.
• This technique holds great importance for the development of optical measurement and fabrication technologies.