Research On Key Techniques Of Fringe Projection Profilometry

Fringe pattern profilometry measures three-dimensional profile data of object by projecting the fringe pattern onto the object and computing the phases of the deformed fringe patterns. Having the advantages of non-contact, high speed, high accuracy, dense data and so on, it has become an important research subject in the three-dimensional sensing in recent years, and has wide application in the fields of industrial manufacture, medical equipment, computer vision,cultural heritage protection and studio entertainment. Although considerable progress has been made in many application fields for fringe pattern profilometry, there are some problems in phase unwrapping, system calibtation and optimization, and three dimensional realtime measurement, which need to resolve.In this dissertation, the phase unwrapping, the system calibration, the calibration of camera lens distortion, and the automatic selection of cutoff frequencies of bandpass filter for the Fourier transformation profilometry (FTP), which are among the key techniques of fringe pattern profilometry, are studied exensively. The main contributions of the dissertation are as follows:(1) An improved mask cut method for two-dimensional phase unwrapping is proposed. Network flow ideas are taken for the method to build the strategy of mask cut generation, in order to generate thin mask cuts to balance residues.The predecessor index is recorded for every pixel during searching, and the path from the current residue to its last residue ancestor is traced by the predecessor index information. This path is taken as an optimal path between the two residues. All the pixels on these paths compose the final mask cuts. The proposed method successfully prevents most of the unnecessary pixels from going into the mask cuts. A significant improvement on the unwrapping result is achieved by the proposed method. The experimental results show that, comparing with the original mask cut method, the proposed method generates thinner and more accurate mask cuts than that generated by the original method.(2) For system calibration, a new phase-to-height relationship for arbitrary optical arrangements in fringe projection profilometry is presented. Based on photogrammetry theory, the unified formulas of height to absolute phase are derived for the reference plane and object surface, respectively. Then the phase-to-height relationship is derived by the subtraction of the reference's phase distribution from the object's phase distribution. Calibration procedures are proposed to determine the parameters required in the relationship for 3D shape measurement. The proposed method is rigorously accurate to compute height from phase on conditions that lens distortions exist on camera lens and projector lens. Compared with some existing methods, the proposed method is not only suitable for arbitrary optical arrangements, but also able to cancel the effect of lens distortion effectively. Simulation results show that the proposed method makes considerable improvement in accuracy.(3) For the calibration of camera lens distortion, a method for computing a complete camera lens distortion model is proposed. With a view of a planar pattern, an iterative procedure is established to estimate the ideal homography under radial, tangential and prism distortions, by specially selected parameters of both radial and tangential distortions. The ideal image points which represent the correct positions of distorted image points can be computed with the approximate ideal homography. Further, the distortion models can be robustly computed with both sets of points, i.e., the distorted points and the corrected ones. Experimental results show that the proposed method is easy to use, and works with high presion in lens distortion correction.(4) For the Fourier transformation profilometry(FTP) which is one of the fringe pattern profilometries, an automatic selection method for parameters of bandpass filter in Fourier transformation profilometry is presented. In this method, with only one frame of fringe pattern, the numerical differentiation and image histogram techniques are applied to FTP to automatically obtain the center and bandwidth of fundemental frequency, and to realize the automatic selection for filter parameters which are used for the automatic processing of frequency domain filtering. The proposed method is adaptive to the system configuration and measurement environment, and can exploit the dynamic measurement advantages of FTP to the full and make the FTP to realize non-stop realtime on-line measurement, therefore has wide practical applications in engineering.