Phase Calculation Of Fringe Structured Light Based On Deep Learning

Three-dimensional(3D)measurement with digital fringe projection(DFP)is one of the most widely used technologies own to the advantages of non-contact,high resolution and high precision,etc.With the increasing demand for 3D applications,higher claims are placed on 3D measurement technology of DFP in various industries,especially in terms of efficiency and accuracy.In recent years,deep learning has developed rapidly and is widely used in the fields of intelligent manufacturing,autonomous driving,and computer vision,which brings inspiration to solve the traditional problem in 3D measurement with DFP by using deep learning.The phase error caused by the nonlinear response between the camera and the projector is one of the important factors that affect the 3D measurement accuracy of DFP.Generally,the most effective way to solve this problem is increasing the number of phase shift steps,however,it will be time-consuming due to the projection of multi-frame fringe patterns.Therefore,in this paper,we propose a nonlinear phase error correction method based on deep learning.By analyzing the information and features of the three-step phase shift pattern,and using the twelve-step phase shift pattern as the label,a multi-level constrained learning network structure is established.We obtained a large number of data sets to train the network to determine its parameters.It is verified that the training network realizes fast nonlinear phase error correction,which can reduce the phase error by 93%,and has high robustness and universality.To obtain high-precision phase with a single fringe pattern is of great significance to improve the efficiency of 3D measurement with DFP.In this paper,we analyzed the relationship between the fringe pattern and the wrapped phase,combined with deep learning to construct a single fringe pattern and wrapped phase multi-level feature joint network structure based on UNet,which realizes the high precision phase acquisition with single fringe pattern.Experimental results show that the phase accuracy obtained by this method is higher than that of classical Fourier fringe profilometry.