Computational 3D Color Imaging And Measurement With Structured Illumination

Structured illumination-based computational three-dimensional(3D)color imaging and measurement,with advantages of non-contact,high resolution,high precision,high fidelity,and high speed,is one of the most representative technologies in the field of optical 3D imaging and measurement,which has been widely used in industrial inspection,cultural heritage,medical cosmetology,entertainment consumption,etc.This dissertation focuses on the research associated with the nonlinear response,hybrid coding,color texture reconstruction,and automatic complete 3D digitization in structured illumination-based computational 3D color imaging and measurement.The goal is to carry out universal nonlinear phase error modeling and suppression,fast hybrid-coding 3D reconstruction using phase-shifting and random fringe patterns,high-accuracy texture mapping and high-quality texture fusion in color texture reconstruction,and robot-assisted automated 3D reconstruction,solving problems of accuracy,speed,color,and automation in 3D imaging and measurement.In structured illumination-based 3D measurement,the phase error causing by the nonlinear response of the digital projector–camera setup is one of the major error sources that reduce the measurement precision.Most of methods for compensating the nonlinear phase error require some additional auxiliary conditions,such as phase benchmark building,gamma calibration,and response curve fitting,etc.,which suffer from the lack of flexibility and robustness.Alternatively,the use of Hilbert transform in fringe analysis can adaptively compensate the nonlinear phase error without additional auxiliary conditions.Nevertheless,the reflectivity jump or phase truncation on the measured surface may result in frequency aliasing or spectrum leakage during Hilbert transform.This dissertation analyzed those problems in theory and proposed a novel Hilbert transform-based method to suppress the nonlinear phase error.Strict mathematical derivation and experimental results demonstrated that the proposed method can effectively overcome the frequency aliasing and spectrum leakage in Hilbert transform caused by non-smooth reflectivity and fractional period fringe,thus suppressing the nonlinear phase error without additional auxiliary conditions and improving the universality and reliability of the proposed method.Phase extraction and unwrapping are two decoding steps in the 3D reconstruction with structured illumination.In general,a wrapped phase map can be accurately,robustly unwrapped with additional coding patterns whereas it increases the time cost of projection and capture,and thus reduces the measurement speed.This dissertation proposed a fast hybrid-coding 3D reconstruction method using sinusoidal and random fringe patterns.Only one random fringe pattern is needed to realize fringe order matching and homologous point searching without the use of phase unwrapping.Through developing algorithms of pattern generation and gray correction of the locally weak-correlated random fringe and fringe order matching of the wrapped phase,the accuracy,robustness,and universality of the hybrid-coding 3D reconstruction are further improved.The performance of the proposed method is consistent with that of the classical gray-code phase unwrapping method in 3D reconstruction accuracy of object surfaces with different materials.However,the gray code algorithm usually requires 7 additional patterns for phase unwrapping,whereas the proposed method only requires one random fringe pattern.Texture mapping and texture fusion are two key techniques for realistic texture reconstruction.The determination of texture parameters directly influence the accuracy of texture mapping and the quality of texture fusion.This dissertation proposed a highaccuracy landmark-aided texture mapping algorithm.The spatial coordinates of the coded landmarks and the parameters of the texture camera are simultaneously obtained with aided of the photogrammetry,and then the parameters of camera orientations are optimized using the nearest point iteration algorithm,which can effectively improve the accuracy of texture mapping.In the process of texture fusion,the alignment operation of global target image is carried out by combining the composite weight and bidirectional similarity function,and then the energy function is optimized by two-step iterative algorithm to reduce the texture dislocation gradually,and overcome the color jump and the texture blur.Finally,the affine invariance property of barycentric coordinates is used to realize the fast automatic transformation of the texture map in color 3D model under arbitrary mesh parameterization relations.The core to provide an automatic 3D scanning of unknown objects is to address the issue of the view planning.The challenge lies in automatically generating the globally optimized scanning viewpoints and performing complete 3D scanning with the minimum number of viewpoints according to the information on geometrical size,topological structure,and surface texture,etc.,with respect to the unknown objects.To this end,an automatic 3D scanning method for the unknown objects is proposed.A rough 3D model of the measured object is first acquired by an infrared depth camera,and then an energy field of viewpoints can be computed according to the surface information of the rough model and the internal constraint of the 3D sensor.Finally,globally optimized scanning viewpoints are generated from the energy field distribution.Furthermore,a two-step calibration strategy for optical measurement system is designed for automatic and complete 3D digitization.Experimental results demonstrated the effectiveness of the proposed method.