Prof. Chao Zuo

School of Electronics and Optical Engineering, Nanjing University of Science and Technology, China

Research Areas:  Computational Optical Microscopy, Optical 3D Metrology, Phase Retrieval, and Digital Holography

Speech title: Deep learning enabled high-speed 3D imaging based on fringe projection profilometry

Abstract

Deep learning is transforming a range of disciplines and eclipsing the state-of-the-art achieved by earlier machine-learning techniques. It has created new opportunities to revolutionize optical metrology techniques. In this talk, we introduce our recent efforts to apply deep-learning approaches to fringe projection profilometry (FPP). We show that the deep-learning-enabled fringe analysis approach can significantly boost the accuracy and improve the quality of the phase reconstruction compared to conventional Fourier transform and windowed Fourier transform approaches. Deep learning can also be used to perform phase unwrapping and outperform conventional multi-frequency temporal phase unwrapping in terms of both unwrapping reliability and robustness.  As a result, with the aid of deep learning, we can use less or even a single raw image for absolute phase retrieval, which enables FPP techniques to go a step further in high-speed and high-accuracy 3D surface imaging of transient events.