The Research On Industrial Application Of Imaging And Stream Computing

In industrial automation production process, target recognition、measurement andcontrol robots based on machine vision system now is everywhere. These robots aremore efficient in mass production、product quality control, and they make great profit.However, some common problems in machine vision system are unsolved for a longtime, including imaging system design、high speed image communication、real-timeimage processing and effective algorithm, which have restricted its application area.This thesis aims to solve these problems and build a universal architecture formachine system design. In this thesis, the core technology in machine vision systemare investigated, including industrial method、device、sign al processing circuit forimaging system, as well as parallel and real-time image computing method formachine vision algorithm. As to specific application, this thesis also research on theimaging system design、image computing system design and the way to coordinatemechanical motion、electrical controller and machine vision system for real-timetarget measurement and control in product process.Chapter One introduces the background and the importance of this thesis.Through practical machine vision system application, the importance of industrialimaging and image computing system is revealed. The research contents of this thesisand the relative technology are introduced briefly, including imaging device and itshistory、the development of image processor. New trends in imaging system are alsointroduced and intelligent imaging system architecture is analyzed. At the end, theresearch strategy and outline of the thesis are given.Chapter Two is mainly on imaging system. First the principals of industrialrobots used for measurement and control based on machine vision system areintroduced. Especially, their requirements on imaging system are analyzed. With theaim to meet these requirements, imaging methods are analyzed from the perspectiveof light physical properties and specific applications. The core of industrial imaging,solid imaging technology is explained in great detail, including photo-sensitiveelement、imaging device such as CCD, CMOS active image sensor. At the end of thechapter, active pixel sensor is modeled and analyzed thoroughly with micro-electronicsimulation software. Details of the signal transformation in photo-sensing and readoutprocess are given as the experiment results. This chapter provides theoretical guidance for industrial imaging system design.Chapter Three is mainly on parallel image computing system. In the beginning,traits of image computing and the shortage of serial computing model in traditionalPC are explained. As to the need of high performance image processor, this paperadopts graphic processing unit (GPU), which can parallelize the image computingprocess. The hardware architecture and computing model of GPU are introduced. Tomeet the requirements of real-time machine vision algorithm execution, this chapterpresents the details of most widely used algorithm parallelization method and thismethod is successfully used in an ampoule product multi-station quality control robot.Experiment results verify the effectiveness of the proposed computing method.Chapter Four describes a quality control robot used in product line based onmachine vision system. Its imaging system, electrical controller, quality controldiagram and machine vision algorithm used are given in great detail. The adoptedpositioning method and multi-region measurement algorithm are explained.Experiment results and the working condition of the robot demonstrate the proposedmethod meet the quality control requirements.In the end, the research work in this thesis is summarized, and further researchpoints are recommended.