Research On Defocused Fringe Projection 3D Measurement Method Based On Deep Learnin

With the continuous advancement of industrialization,how to obtain threedimensional(3D)data of objects quickly and accurately has been a research hotspot of researchers’ attention.As an active 3D measurement technique,fringe projection profilometry(FPP)is widely used in industrial inspection,biomedicine,reverse engineering and other fields because of its merits of high accuracy,high efficiency and non-contact.However,with the continuous development of various fields,the increasing demands on the measurement efficiency of FPP have raised higher requirements.The defocusing projection technology only projects 0 and 1 grayscale,thus greatly improving the fringe projection speed while avoiding the gamma effect of the projector,but the traditional defocusing projection 3D measurement technology is difficult to completely remove the higher harmonics in the binary encoded fringe,and there are often large ripple-like errors in the actual measurement,which greatly limits the application of defocusing projection 3D measurement technology in practical production and life.To this end,this paper investigates the problem of high harmonics in binary fringe encoding from the perspective of frequency domain by starting from the mathematical model of defocusing projection,and combines deep learning technology,defocusing projection technology and fringe projection 3D measurement technology to realize a high-precision and high-efficiency 3D measurement method.The main research contents of this paper are as follows.(1)An defocusing fringe projection 3D measurement technique based on deep learning is proposed.The traditional defocusing projection 3D measurement technique has a large measurement error because it is difficult to remove the higher harmonics in the square wave fringe.In this paper,by introducing deep learning into the defocusing fringe projection 3D measurement technique,the data-driven feature of deep learning and its powerful feature learning ability and feature expression ability are used to double the measurement accuracy of the three-step phase shift algorithm of square wave fringe without changing the environmental conditions and hardware parameters,and this method solves the problem that the influence of high harmonics in the traditional defocusing fringe projection 3D measurement technique is difficult to be removed,and greatly improves its measurement accuracy.This method solves the problem of difficult to remove the influence of high harmonics in the traditional defocusing fringe projection 3D measurement technology,greatly improves its measurement accuracy,and realizes a 3D measurement method with both high accuracy and high efficiency.(2)Through the mathematical modeling analysis of the defocusing effect,we find that although several widely used binary coding methods "push" the high harmonics away from the fundamental frequency in the frequency domain by various methods,the high harmonics are still difficult to be removed by the slight defocusing effect when the fringe frequency is low,which makes the traditional defocusing it difficult to use the traditional defocusing 3D measurement technique in the low frequency domain.We improved the existing network structure,and after a lot of targeted training,we finally improved the measurement accuracy of low-frequency square wave fringe three-step phase shift method by 3-7 times without changing the environmental conditions and hardware parameters,which solved the problem that the traditional defocusing fringe projection 3D measurement technology is difficult to be applied in the low-frequency field,and provided a new idea for the subsequent real-time high-precision3 D measurement technology.