Single-shot 3D measurement via deep learning fringe projection profilometry with geometric constraints

Abstract

Single-shot three-dimensional (3D) measurement has always been the ultimate goal of fringe projection profilometry (FPP). Some studies have shown that deep learning outperforms traditional algorithm in analyzing single fringe pattern for complex scenarios. However, accurately phase unwrapping for a single wrapped phase map remains a significant challenge. In this paper, we propose a deep learning-based fringe projection profilometry. This method considers the geometric constraints of the measurement system. With the reference phase generated by the calibration parameters and appropriately designed intermediate variables based on physical models and prior knowledge, the proposed method is capable of recovering high-quality absolute phase from a single fringe pattern at the accuracy sufficiently high to rival traditional multi-frame algorithms. In addition, as far as FPP is concerned, the significance of the reference phase generated by the calibration parameters of the measurement system for deep learning-based single-frame phase unwrapping is experimentally demonstrated. Experiments on both static and dynamic scenarios show that the proposed method can achieves motion-artifact-free and high-resolution single-shot 3D measurements in various complex scenarios using only a single-frequency fringe projection.