Deep Learning-Based Single-Frame Depth Reconstruction Empowered by Indirect Mapping

The precise and efficient 3-D measurement has always been the key research focus in fringe projection profilometry (FPP), and achieving high-precision 3-D measurement using a single pattern is the ultimate goal pursued in this field. In recent years, deep learning-based fringe pattern analysis methods have made significant progress, and based on this, we propose a deep learning-based single-frame FPP. This method trains two parallel U-Nets to restore the wrapped phase and the depth information, which serve as intermediate variables. Moreover, it combines the phase-to-depth model and absolute phase renormalization operation for subsequent calculation and phase correction, ensuring the accuracy of the depth information. Experimental results demonstrate that the proposed method can achieve high-precision depth reconstruction from a single fringe pattern in complex and dynamic scenarios, which effectively overcomes the limitations of conventional methods and significantly improves the accuracy and robustness of measurement. The method offers a viable solution for achieving high-precision depth reconstruction using single-frame techniques.