Research On Three-Dimensional Reconstruction Of Transparent Target Based On Polarization Analysis

The three-dimensional reconstruction of the object surface is an important subject in the fields of optical engineering and computer graphics.It has a good application and research value in many fields such as archeology,industrial manufacturing,virtual reality,aerospace,medical,and visual inspection.Among many kinds of measured targets,transparent targets have always been a difficult task for three-dimensional reconstruction.Due to the low reflectivity,the complexity of internal light transmission,the susceptibility to the surrounding environment,and the lack of texture details on the surface,it is difficult to achieve three-dimensional reconstruction of transparent surfaces with traditional methods.Therefore,the research on three-dimensional reconstruction of transparent targets has important theoretical significance and practical value.This paper presents a method for three-dimensional reconstruction of transparent targets based on polarization analysis.The measurement system of single-side circularly polarized light is used to extract the normal and gradient information of the targets,and it is combined with a high-precision gradient integration algorithm to complete the calculation of height data and three-dimensional reconstruction of shape.Firstly,in a circularly polarized light environment,the polarization state of the light reflected off a transparent surface is studied and discussed.According to the propagation process of polarized light on the interface,a mathematical model about the relationship between the degree of circular polarization and the normal vector is derived,and the ambiguity solution of the key parameters is eliminated by constraining the position structure of the measurement system.Secondly,according to the approximate orthogonalization method of the pinhole model,the reference plane of the measured surface is determined and a spatial coordinate system is established.Combining the camera calibration technology is to determine the conversion relationship between this coordinate system and the pixel coordinate system.The projection process from the pixel plane to the reference plane is studied.Thirdly,the basic principle and mathematical model of the Fourier-based modal method are analyzed,and its difference sampling model is improved based on Taylor series theory to reduce reconstruction errors.In order to make the calculated data of the new model to meet the requirements about periodicity and size in the discrete Fourier transform,the original data is extended by combining the antisymmetric extension and the period extension,and Gerchberg iteration theory is introduced to reconstruct non-rectangular data.This paper validates the performance of the proposed algorithm in terms of accuracy and calculation time through simulation experiments.The results show that the accuracy is significantly improved compared to traditional algorithms,and the calculation speed is fast.It is suitable for reconstruction of large-scale data.Finally,the feasibility and effectiveness of the proposed algorithm are verified by the experiment.The experimental results show that the reconstructed shape is highly consistent with the original surface,and its accuracy is high.For a surface with a radius of 98.97 mm,its root mean square error is about 0.018mm;for the high-order surface in the paper,its root mean square error is about 0.016 mm.