Structured-light 3D shape measurements using deep learning

Abstract

Over the past few decades, tremendous efforts have been devoted to developing various techniques for fringe analysis, and they can be broadly classified into two categories: (1) phase-shifting (PS) methods which require multiple fringe patterns to extract phase information and (2) spatial phase demodulation methods which allow phase retrieval from a single fringe pattern, such as the Fourier transform (FT), windowed Fourier transform (WFT), and wavelet transform (WT) methods. Compared with spatial phase demodulation methods, the multiple-shot phase-shifting techniques are generally more robust and can achieve pixel-wise phase measurement with higher resolution and accuracy. Furthermore, the phase-shifting measurements are quite insensitive to non-uniform background intensity and fringe modulation. Nevertheless, due to their multi-shot nature, these methods are difficult to apply to dynamic measurements and are more susceptible to external disturbance and vibrations. Thus, for many applications, phase extraction from a single fringe pattern is desired, which falls under the purview of spatial fringe analysis. Recently, we demonstrated that the use of convolutional neural networks can substantially enhance the accuracy of phase demodulation from a single fringe pattern. Moreover, we find the powerful learning ability of deep neural network (DNN) enables the phase unwrapping, super-fast 3D shape measurement of transient events, multi-view fringe projection and so on. From comparative results, the DNN shows better performance over traditional state-of-the-art methods in terms of the phase accuracy and efficiency. We believe the deep learning technique is a powerful technique to handle fringe images and will find wide applications in 3D measurements with structured-light illumination.