Study on gamma correction for three-dimensional fringe projection measurement based on attention U-Net network

The gamma distortion of a structured light system significantly reduces the measurement accuracy. A gamma correction method based on an attention U-Net deep neural network is proposed in this work. The mapping relationship between the wrapped phase obtained by a three-step phase-shifting algorithm and the wrapped phase obtained by a 12-step phase-shifting algorithm is established by the learning capability of the attention U-Net network model. Pixel-wise error correction of the wrapped phase maps is enabled by this method. To validate the effectiveness of the proposed method, the relationship between the number of phase-shifting steps and the nonlinear phase error was analyzed experimentally. Phase correction and 3D point cloud reconstruction experiments were carried out. It was proved that the proposed method eliminates the need to analyze the complex mathematical model of the gamma distortion, which reduces the nonlinear phase error by three times without increasing the number of fringe frames. Compared with the traditional three-step phase-shifting algorithm and the active pre-correction method, the 3D reconstruction accuracy was effectively improved by the proposed method.