Research On Phase Retrieval Of Fringe Projection Profilometry 3D Measurement System Based On Deep Learning

Fringe projection profilometry(FPP)measurement technique has been widely used in many fields such as intelligent manufacturing,virtual reality,heritage protection and medical diagnosis because of its characteristics such as non-contact,simple equipment,high speed,high accuracy,strong robustness to ambient light and so on.FPP measurement technique based on phase-shifting method needs a large number of projection patterns,which is often used in the static 3D measurement system.But it will encounter many difficulties in the dynamic scene.Therefore,how to reduce the number of projection patterns while maintaining the accuracy has become a research hotspot in recent years.At the same time,the vigorous development of deep learning and convolutional neural network also provides a new scheme to reduce the number of projection patterns.The main content of this paper is to apply convolutional neural network to reduce the number of projection pattern frames.The main innovations and work are as follows:(1)In this paper,a method based on convolution neural network is proposed to obtain the absolute phase of dual-frame patterns.In this method,fringe pattern and speckle pattern are projected.The fringe pattern is used to obtain the wrapped phase by convolution neural network and the solving strategy of obtaining wrapped phase is analyzed in detail.Then,the fringe order is obtained by convolution neural network based on the high spatial resolution of speckle pattern.In this method,the speckle pattern is binarized to adapt to the task of phase unwrapping.At the same time,a method is provided to correct the absolute phase error caused by the mismatch between the wrapped phase and fringe order.The experimental results show that this method can obtain high precision absolute phase only from double frame measured scene patterns.(2)In this paper,a method of absolute phase retrieval using single frame fringe pattern is proposed,which further reduces the number of projection pattern.In this method,the convolution neural network structure in(1)is still used to obtain the wrapped phase,and a new phase unwrapping method is proposed.Since it is difficult to extract the feature information from a single frame fringe pattern to obtain the fringe order,phase unwrapping is assisted by reference plane fringe order information.In this method,the reference plane only needs to be pre-shot once.In the actual measurement process,there is no need to repeat the shooting,so it will not affect the measurement speed.Experimental results show that this method can effectively recover the absolute phase in dynamic scenes.(3)Finally,a single frame FPP measurement system is built in this paper and the absolute phase obtained from a single frame fringe pattern combining with calibration parameters are used to realize 3D reconstruction.The experimental results show that the proposed a single frame absolute phase retrieval method can be successfully applied to the FPP measurement system,which provides a feasible reference scheme for dynamic 3D reconstruction task.