Fig. 1. Traditional structured light 3D measurement system

Fig. 2. Principle of phase unwrapping in Gray code

Fig. 3. Convolutional neural network structure

Fig. 4. Three modules of a neural network. (a) Multi Block module; (b) Res Path module; (c) Third module

Fig. 5. Global guidance path network structure

Fig. 6. Training dataset

Fig. 7. Depth and error images of an object. (a)(f) Depth map; (b)(g) predicted 3D shape images predicted by U-Net; (d)(i) predicted 3D shape images predicted by MultiResHNet; (c)(h) error images measured by U-Net; (e)(j) error images measured by MultiResHNet

Fig. 8. Noised training dataset

Fig. 9. 3D effect and error images of an object. (a) 3D shape image corresponding to a fringe pattern; (b) 3D shape image predicted by U-Net; (d) 3D shape image predicted by MultiResHNet; (c) error image measured by U-Net; (e) error image measured by MultiResHNet

Fig. 10. Neural network training dataset

Fig. 11. RMSE of neural networks. (a) U-Net; (b) MultiResHNet

Fig. 12. Visual representation of three-dimensional reconstruction of objects. (a)(e) Fringe patterns of network input; (b)(f) 3D shape images corresponding to fringe patterns; (c)(g) 3D shape images predicted by U-Net; (d)(h) 3D shape images predicted by MultiResHNet

Fig. 13. Error results for multiple objects. (a) Fringe patterns of network input; (b) error images predicted by U-Net; (c) error images predicted by MultiResHNet