Fast-speed Interferometry Based On CUDA And OpenCV

Among optical detection technology,phase shifting interferometry(PSI)can realize the non-contact measurement of the wavelength,which is one of the most effective methods in the optical detection.The algorithm of multi-step phase-shifting is widely used for its high accuracy and repeatability,but it has a large amount of data and high computational complexity.In addition,CUDA is a framework of heterogeneous parallel computing technology,which can effectively solve the common problems in scientific computing.Besides,OpenCV is the open source computer vision library,provides an effective solution to the problem of image processing.Therefore,to solve the bottleneck problem in phase shifting interferometry,the thesis puts forward the project of ‘Fast Interferometry based on CUDA and OpenCV ‘.The foundation of phase-shifting interferometry has been introduced,including the principle of Fizeau interferometer,some typical phase-shifting algorithms and the basics of phase unwrapping algorithm.Besides,the phase shifting error of interference measurement has also been carefully discussed.A simulated analysis on anti-noise performance of four classical phase-shifting algorithms has been conducted.The result shows that the stochastic phase shifting(RPSI)algorithm has better anti-noise performance and faster than the fixed step phase-shift algorithm.Secondly,foundation of CUDA programming has also been discussed,including its programming model,implementation model,storage model and the syntax of CUDA.Then the pretreatment process of interferogram has been discussed.A special mask is used to eliminate the calculated independent region,and the total computation is reduced by 1/5.There are two methods to create the special mask in OpenCV.The first one is based on the Hough transform circle detection and the other is a segmentation method based on contour region.The result shows that the method based on the principle of Hough transform can only solve the problem that the outline of central fringe area is a standard circle.While the latter one can effectively extract the approximate circular interference fringe area.Finally,the problem of fast interferometry measurement based on CUDA is discussed in detail.The whole design process of the algorithm is discussed systematically.It discusses the parallelism analysis and task decomposition of the algorithm,and the details of the parallel algorithm and the result analysis.GPU-accelerated AIA random phase shift algorithm,GPU-accelerated Goldstein phase unwrapping algorithm,and the use of least squares method to remove the tilt in shape surface.The whole process includes reading 13 frames from the interferogram video stream,preprocessing,solving the wrapping phase,removing the tilt in the solved shape surface,and finally obtaining the surface topography of the component under test.Among them,an average of 176 milliseconds is consumed to solve the phase distribution of the 13-frame interferogram,averaging 34 milliseconds to complete the shape tilt-removing.For a single overall measurement process,it costs total 480 milliseconds to computing the shape surface of the tested optical part,and individually,176 milliseconds for solving the wrapped phase distribution from 13 interferogram in a video stream,34 milliseconds for phase unwrapping on average.Overall,the whole computing process of interferograms achieved with a reasonable performance which can fulfill the demand of fasting phase shifting interferometry.