基于GPU的数字全息实时再现技术研究

发布时间：2018-03-08 00:14

本文选题：数字全息　切入点：数字全息粒子追踪测速　出处：《重庆理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：数字全息技术是一种利用图像传感器记录物体的振幅和相位信息,并通过计算机模拟光学的衍射过程而进行物体再现的技术。数字全息测量技术具有无接触,无损伤等优点,可以用于物体的三维形貌、应力应变场、温度场、流场等领域的观测。在一些测量领域,需要实时测量并再现测量目标的动态变化过程并且同时对目标场进行三维空间的实时再现,数据计算量极大,因而必须要加快数字全息图处理的速度。本文将基于GPU(Graphics Processing Unit,图形处理器)的并行计算技术引入到DHPTV(Digital Holographic Particles Tracing Velocimetry,数字全息粒子追踪测速)中,数字全息图记录了粒子的三维空间信息,通过对数字全息图进行重建,提取粒子的三维空间坐标信息,为了能够清晰观测不同深度信息的粒子,需要对整个粒子场以一定的重建间隔进行全场重建,然后通过粒子匹配,可以获得粒子的三维速度矢量场。由于全场重建的计算量极大,采用了基于GPU的并行计算技术极大提高了重建速度,实现对空间粒子场的三维速度矢量场的实时重建。为了实现粒子场三维速度矢量场的实时再现,本文主要进行了如下研究:(1)数字全息基本原理以及数字全息图重建算法研究;(2)研究了DHPTV的测速原理,提出自己的算法以及计算流程,通过对两幅旋转粒子的全息图进行重建测速,对测速结果进行误差分析,验证实验方法的可行性以及结果的精度;(3)设计了一种数字全息实时显微镜,该显微镜主要分为硬件系统和软件系统,硬件系统主要是该系统所使用的硬件设备以及光学元件,软件系统主要是基于GPU的并行算法程序。软件系统的设计思路是利用Matlab中丰富的API(Application Programming Interface,应用程序编程接口)函数库建立计算模型,利用CUDA(Compute Unified Device Architecture,统一设备计算架构)扩展C++语言对算术均值滤波,连通域识别、图像归一化以及粒子匹配等算法进行编程。为了实现GPU利用率的最大化,使用Kepler架构GPU所具有的Hyper-Queue(超工作队列)特性进行多流编程。对比了Matlab程序运行结果以及Matlab调用CUDA程序运行结果,得到加速比。
[Abstract]:Digital holography is a kind of technology which uses image sensor to record the amplitude and phase information of object and reproduce the object by simulating the diffraction process of optics by computer. The digital holographic measurement technology has the advantages of no contact, no damage and so on. It can be used to observe three-dimensional topography, stress-strain field, temperature field, flow field and so on. It is necessary to measure and reproduce the dynamic change process of the measurement object in real time and reproduce the target field in three dimensional space at the same time. Therefore, we must speed up the processing of digital holograms. In this paper, the parallel computing technology based on GPU(Graphics Processing unit is introduced into DHPTV(Digital Holographic Particles Tracing velocimetric (digital holographic particle tracking velocimetry). The digital hologram records the three-dimensional spatial information of the particle. By reconstructing the digital hologram, the three-dimensional coordinate information of the particle is extracted, so that the particles with different depth information can be clearly observed. It is necessary to reconstruct the whole particle field at a certain interval, and then the three-dimensional velocity vector field of the particle can be obtained by particle matching. The parallel computing technology based on GPU is used to greatly improve the reconstruction speed and realize the real-time reconstruction of the three-dimensional velocity vector field of the space particle field, in order to realize the real-time reconstruction of the three-dimensional velocity vector field of the particle field. In this paper, the basic principles of digital holography and the reconstruction algorithm of digital holograms are studied as follows: the principle of velocity measurement of DHPTV is studied, and its own algorithm and calculation flow are put forward. A digital holographic real time microscope is designed by reconstructing the hologram of two rotating particles and analyzing the error of the velocity measurement results to verify the feasibility of the experimental method and the accuracy of the results. The microscope is mainly divided into hardware system and software system, the hardware system is mainly used in the system of hardware equipment and optical components, The software system is mainly a parallel algorithm program based on GPU. The design idea of the software system is to make use of the rich API(Application Programming interface (application programming interface) function library in Matlab to build the calculation model. Using CUDA(Compute Unified Device Architecture, the extended C language is used to program arithmetic mean filtering, connected domain recognition, image normalization and particle matching. By using Hyper-Queue (Hyper-Queue) feature of Kepler architecture GPU for multi-stream programming, the speedup ratio is obtained by comparing the running results of Matlab programs and the results of Matlab calling CUDA programs.
【学位授予单位】：重庆理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;O438.1

【参考文献】