Single-frame two-stage fringe projection profilometry based on deep learning

For dynamic objects with surface discontinuities, traditional fringe projection profilometry struggles to obtain accurate three-dimensional information. To address this challenge, this paper presents a single-frame, dual-stage fringe projection profilometry technique that requires only one deformed fringe pattern and employs two neural networks. The first neural network predicts deformed fringe patterns at different frequencies, while the second neural network predicts the wrapped phase numerator and denominator for each frequency. By integrating a traditional multi-frequency phase unwrapping method with system calibration, a step-by-step 3D measurement process is achieved. Moreover, this paper introduces a convolutional neural network called DARU-Net, which is based on U-Net and demonstrates significant advantages over U-Net and its derivatives in deep learning tasks. The experimental results show that the proposed method can accurately predict the 3D information of objects with surface height discontinuities using only a single fringe pattern, thus expanding the application scenarios of 3D measurement.