Research On Digital Holography Real-time Reconstruction Technique Based On GPU

Digital holography uses imaging sensor to record phase and intensity of an object and reappear the object through simulating optical diffraction process by computer.Digital holographic measurement is a non-contact and non-destructive measurement technique as same as optical measurement.It can be applied to 3D shape measurement,stress-strain field measurement,temperature field measurement and micro-fluid field measurement.In some areas,it needs real-time reconstruction of dynamic changes of 3D object.The main problem of realizing real-time reconstruction is the massive computation.Therefore,it is imperative to accelerate image processing to synchronize the dynamic changes of object with reconstruction.Parallel computation based on GPU is applied to DHPTV.Digital hologram contains spatial coordinate of 3D object.Spatial coordinate is extracted by reconstruction of digital hologram.In order to obtain clear reconstruction image of particles in varying depth,it needs to reconstruct the digital hologram throughout the full-field of object piece by piece.3D velocity vector field of particles is worked out by particle pairing.Since computation of full-field reconstruction is massive,GPU is adopted to realize real-time reconstruction of3 D velocity vector field of particles.In order to realize real-time reconstruction,essential researches of this paper are as follows:(1)Fundamental principles of digital holography and reconstruction methods are investigated;(2)After analysis of principles of DHPTV,an algorithm is proposed.DHPTV is applied to velocity measurement of rotating particle.The experiment results validate the reliability and accuracy of the method.(3)A real-time digital holographic microscopy is designed specializing in measurement of 3D velocity vector field of particles.The microscopy is consisted of hardware and software.The hardware system is holographic apparatus and optical components.The software system is the parallel algorithm based on GPU.Computational model is established taking advantage of abundant API of Matlab.Algorithms of mean-value filtering,recognizing connected component,normalization,and particles pairing are programming utilized CUDA extended C language.In order to acquire maximum performance of GPU,Hyper-queue technique which is a specific character ofKepler architecture is used to GPU programming.The execution results are compared with serial computation.The speed-up ratio is worked out.