3D Measurement With Large Depth Of Field Structured Light Based On Deep Learnin

Structured light 3D measurement technology has been widely used in many fields because of its fast speed,high accuracy and high precision.However,how to perform 3D reconstruction with high performance in large depth of field scenes has always been one of the difficulties.At present,the research on 3D reconstruction algorithms in large depth of field scenes has its own advantages and disadvantages.In this paper,we propose a deep learning based 3D measurement method for large depth of field,which takes into account both the speed and accuracy of measurement in the process of 3D reconstruction in large depth of field scenes.Aiming at the problem of phase unwrapping in large depth of field,a three-gray level Gray code method is proposed,which can not only obtain better phase unwrapping,but also improve the efficiency of measurement.The specific research work of this paper is as follows:(1)For that problem that high performance 3D reconstruction cannot be carry out in a large depth of field scene at present,a large depth of field 3D measurement method based on deep learn is proposed.Firstly,the influence of camera defocus and projector defocus on 3D measurement accuracy in large depth of field scene is analyzed.Secondly,a convolutional neural network with extended depth of field is built,which is mainly composed of three parts:U-shaped structure,residual structure and layer jump connection.The whole large depth of field scene is divided into each sub-depth of field,and each sub-depth of field does not need to be calibrated respectively,but only needs to be calibrated once at the focusing positions of the camera and the projector.In the network training process,each sub-depth of field is respectively trained to obtain a plurality of different models,and in the testing process,the sub-depth of field of the object to be tested is determined through a plurality of groups of depth maps obtained by the plurality of models,so that a corresponding three-dimensional effect map can be output.Finally,the depth of field expansion ability and error reduction ability of this method are verified.The experimental results show that the proposed method can extend the measurement depth of field by 4.3 times.(2)The phase unwrapping problem in large depth of field scene is analyzed,and a three gray level Gray code method is proposed for phase unwrapping in large depth of field scene.Firstly,the common spatial phase unwrapping algorithms and temporal phase unwrapping algorithms are introduced,and the Gray code method and multi-frequency heterodyne method of temporal phase unwrapping algorithm are introduced.Secondly,the problem of phase unwrapping in large depth of field scene is analyzed.The disadvantage of Gray code method for phase unwrapping in large depth of field scene is that the number of projection fringe frames is too large to measure quickly.The multi-frequency heterodyne method is affected by noise,so it can not get accurate phase unwrapping.Finally,the phase unwrapping in the large depth of field scene is carried out by using the three-gray level gray code method,the accuracy of the method is basically the same as that of the two-gray level gray code method,and the number of projection frames is reduced,so that the method can be applied to the field of rapid measurement in the large depth of field scene,and the application range is wider.(3)A 3D measurement system for large depth of field scene is built.Firstly,the experimental equipment used in the large depth of field 3D measurement experiment is introduced.Secondly,the process of calibration is introduced,including the selection of calibration board,the collection of calibration pictures and the analysis of calibration results.Finally,the experiments of plaster toy and standard ball are carried out in the large depth of field scene.The results show that the surface of plaster toy is very smooth,the details are well restored,and the spherical fitting error of standard ball is 0.02 mm.The accuracy and reliability of the 3D measurement system in the large depth of field scene are proved.