外文翻译 - 基于V4L2嵌入式图像采集的实现

外文原文TheImplementationofEmbeddedImageAcquisitionBasedonV4L2LUYinli,YUHongli,ZHANGPengpengPhysicalElectronics,DepartmentofElectricalEngineeringofUniversityofSouthChinaHengyang,PeoplesRepublicofC,AbstractInviewofthecurrentsituationofthelessembeddedLinuximageacquisitionapplicationbasedonV4L2(VideoforLinuxTwo),embeddedimageacquisitionsystembasedonV4L2isdesigned.S3C6410isadoptedashardwaredevelopmentplatformandLinuxassoftwareplatform.AfterloadingthedriverofCMOScameraOV9650andotherrelativedrivers,imageacquisitionapplicationsonthebasisofV4L2iscompiled,adoptingdoubleframebuffertoreduceCPUutilizationratetoaccomplishacquisitionandpreviewimage.Withthesizeof400480andtheframerateof15framespersecond,thepicturecollectedshowsclearlyandithasgoodquality.Theexperimentresultindicatesthatthesystemhasgoodscalability,anditoffersanewideaforembeddedhand-helddevices.Keywords-V4L2;OV9650;imageacquisition;S3C6410;doubleframebufferI.INTRODUCTIONThevideoimageprocessingtechnologyiswidelyusedinelectronictechnologyandcomputertechnology.Theimagefromacameraorotherimageacquisitionequipmentsistransformedmathematicallytogettheinformationofdigitalimage;then,itisencoded,filtered,enhanced,recovered,compressed,storedandotherprocessesconductedbythecomputer;visualimagesandotherresultsfinallywillbereceived.Withthedevelopmentofembeddedtechnology,thecurrentimageacquisitiontechnologyistransformedfromtheanaloguetothedigital.Amongthem,embeddedimageacquisitionattractsmoreandmorepeoplesattentionforitssuperiorperformance.Embeddedmicroprocessoristhebestchoicefortheimageprocessingsystembecauseofitshighperformance,cost-effective,andlowpower.SamsungS3C6410hasARM11coreandadopts64/32bitinternalbusarchitecture.Italsoincludesmanypowerfulhardwareacceleratorssuchasmotionvideoprocessing,audioprocessandsoon.AnintegratedMFC(MultiFormatcodec)supportsencodinganddecodingofMPEG4/H.263/H.264anddecodingofVC1,soitisveryconvenientforimageprocessing.S3C6410isutilizedasthecoreoftheembeddedhardwareplatform,thepaperdesignsaimagecollectionapplicationsonbaseofV4L2.AfteraddingsupportforOV9650andV4L2tothekernelandtransplantingLinuxoperatingsystemtothehardwareplatform,imageacquisitionapplications,usingdoubleframebuffertoreducepowerconsumption,isdevisedonthebasisofV4L2,ultimately,picturewillbegatheredandshowedontheLCD.II.HARDWARESYSTEMSTRUCTUREThewholeimageacquisitionsystemmainlyincludesCMOScameraOV9650,corecontrollerS3C6410,LCDandothersperipherals.Thesystemhardwarestructureisshowedasfigure1:A.Processor:SamsungS3C6410isbasedonARM1176JZF-Score,anditsmainfrequencyis533MHz/667MHz,includingSIMD(singleinstructionmultipledata)formediaprocessing.Itappliesspecialwaystomakeitsuitableforvideoprocessing.B.Camera:OmniVisionsOV9650isusedhere.Itincorporatesa10bitA/Dconverter,correspondingtodataoutputinterfaceD0:9.Theoutputimagedataformatcanbe10bitrawRGBor8bitRGB/YCbCrthroughinternalDSPprocessing.TheCAMIFusedsupportsITU-BT601/656YCbCr8bitmode.Asaresult,D9:2ofOV9650shouldbeconnectedtothedatainterfaceofCAMIFforS3C6410.C.Peripherals:ThispartmainlyincludesLCDandtouchscreeninterfacecircuit,JTAGdebuggingcircuit,resetcircuit,thepowersupplycircuit,RS232serialinterfacecircuitandsoon.BecauseOV9650registerscontrolprotocolSCCB(serialcameracontrolbus)canbecompatiblewithI2C,soweuseI2Cofback-endplatformtosimulateSCCBtoconfiguretheOV9650internalregisters.III.SOFTWAREDESIGNTherearetwostepstoachievetherealizationofimagecollection:firstly,loadtherelatingdrivertoLinuxkernelonthebottom;secondly,compileimageacquisitionapplicationsonthetoptogetimagedata,whichwillbeshowedonLCDfinally.A.LoadtherelatingDriverBeforecompilingtheapplicationsprocedure,driversrelatedshouldbeloadedatfirst.Becausethekerneladoptedincludesthedrivers,sojustaddcorrelativeconfigurationtothekernel.1)kernelconfiguration:Inthedirectoryofkernel,usethecommandmakemenuconfigforkernelconfiguration:chooseDeviceDriversMultimediadeviceVideoforlinux,loadvideo4linuxmoduletoprovideprogramminginterfaceforvideodevice;chooseDeviceDriversMultimediadevicesVideocaptureadaptersSamsungCameraInterfacedriversExternalCameraOVTOV965x,addthesupportofCMOSimagesensortothekernel.Savetheconfigurationandquit.2)generatekernelimageUsethecommandsbelow:makedep:readtheconfigurationfileandcreatdependencytreecorrespondingtotheconfigurationtodeterminecompileoptions;makezImage:generatethekernelimagethatcansupportV4L2andOV9650.Whenthekernelstarts,ifthereisthesameinformationasfollows,itindicatesthatthedriverofOV9650hasbeenloadedsuccessfully:CAMOV9650initregstart.CAMOV9650initregend.Afterthesystemstarts,itcanautomaticallyidentifyOV9650.Systemcallscanbeutilizedtoobtainequipmentinformationtofacilitateapplicationdevelopment,theI/OoperationinterfacefunctionprovidedbyCMOSimagesensordrivercanrealizetherelevantoperation.B.ImageAcquisitionApplicationsV4LisarangeofinterfacefunctionsprovidedbyLinuxforvideodevicesapplications,theimagedatacanbereadoutfromvideoequipmentbyusingAPIfunctions.1)V4L2driverstructureV4Lhastwoshortages:whendrivingprogramisbeingdesigned,youcannotopenmultipleequipmentatthesametime;furthermore,V4LAPIcannotwellsupportequipmentswithencodingability.Underthiscircumstance,BillDirksredesignedasetofAPIsanddatastructure,calledV4L2.ComparedtoV4L,theextensibilityandflexibilityareimproved,morevideodevicesareincludedinitsmanagement.AlthoughV4L2istheupdatedversionofV4L,theyarenotcompatible.V4L2isappliedasLinux2.6kerneldriveragreement.AsaspecificationsofvideocapturedevicedriverforLinux,V4L2adoptsalayeredapproachtoofferclearandconsistentinterfacefordriverdevelopment:theapplicationisatthetop,V4L2inthemiddlelayer,whiletheactualhardwaredeviceonthebottom.V4L2hastwo-tierdriverarchitecture.Theupperisvideodevmodule.Whenthevideodevhasbeeninitialized,itregistersitselfasacharacterdeviceofmajornumber81,andtheirowncharactersdrivermemberfunctionsareregisteredsimultaneously.ThelowerisV4L2driver,videodevclientinfact,whichcanbecalledbyVideodevthroughV4L2drivermemberfunctions.AfterV4L2driverisinitialized,thestructureincludingV4L2drivermemberfunctions,minornumberandotherrelatedinformationispassedtovideodev,consequently,thedevicetobehandledisregisteredtovideodev.WhenV4L2driverisbeinginitialized,firstofall,systemdevicetobeprocessedisenumerated;then,thestructurev4l2_deviceisaddedforeachdevice,andthepointerdirectingtothestructurev4l2_deviceispassedtov4l2_register_device(),whichcallsinitializationfunctionsofv4l2_devicetoinitializedevices.Beforecallingv4l2_register_device(),theinformationofname,type,minornumberandopenfunctionmustbeaddedtodriver.Whenadriveristouchedoffbyapplications,thecontrolwillfirstlybedeliveredtothefunctionsofvideodev,videodevhasresponsibilitytotransformfileornodestructurepointerpassedbyapplicationstoV4L2structurepointer,andcallsprocessingfunctionsofV4L2driver.2)ImageacquisitionbasedonV4L2ThemainstepsofvideocaptureofV4L2areasfollows:a)openvideodevicev4l2_fd=open(V4L2_DEV_NODE,O_RDWR);open/dev/video0;fb_fd=open(FB_DEV_NODE,O_RDWR);openframbuffer;b)deviceinitializationQuerythedevicepropertiesandgetfunctionsupportedbyVIDIOC_QUERYCAP;setpictureformatusingVIDIOC_S_FMT;allocatememorybufferinthekernelspaceanduserspacerespectivelythroughVIDIOC_REQBUFSandmalloc(),andfinallyutilizemmap()formemorymapping.c)imageacquisitionanddisplayStartandstopacquisitionbyVIDIOC_STREAMONandVIDIOC_STREAMOFF.Therearetwowaysforvideoacquisition:memorymappingandreadingdatadirectlyfromdevice.Imagecollectionbydoublecachethroughmmap()isusedhere.Compilepreviewfunctionv4l2_show_on_fb(v4l2_fd,fbmem,preview_frames)topreviewthepictures.d)closevideodevice3)Closethedevicethroughclose().TheflowchartofimageacquisitionbasedonV4L2isasfigure2:TherearetwoimportantsystemcallsforV4L2duringvideoacquisition:ioctl()andmmap().a)mainioctl()commandioctl()isapowerfulfunction.ItcancontroltheI/Ochannelsofequipment,settheformatofvideoandframe,andalsocaninquirycurrentdeviceproperties.MainioctlcommandsareshownastableI:Headerfilessuchasinclude/linux/videodev2.horvideodev.hcanbeaddedtothekerneltoachievetheabovefunctioncalls.b)memorymappingmmap()Therearetwowaystocapturethevideoimage:oneisread()forreadingdirectly,andtheotherismmap()formemorymapping.Mmap()givesachanceforsharingmemorybetweenprocessesthroughmappingacommonfile.Afterageneralfileismappedtotheaddressspaceofprocess,thefilecanbeaccessedbyprocessesasordinarymemory,thatis,thememorycanbereadandwrotedirectly,andthereisnoneedtocallread()andwrite(),itcanaccelerateI/Oaccess.Inthethirdstepofimageacquisitionabove,datacopyisreplacedbymultiplecacheforimagecollectionthroughmmap()toimproveefficiency.Twoimagebuffersarecreatedinthekernelspace,whenapplicationsneedimages,datawillbesharedbymappingthekernelbuffertouserspacethroughmmap().Theprocessofvideoimagecollectionusingdoubleframebufferisasfollows:First,requestimagebufferthroughVIDIOC_REQBUFS;second,callmmap()twicetocreatecorrespondingrelationbetweenuserspaceandkernelspacebuffer;third,addthebuffertocollectionsequencebesidedriverbyVIDIOC_QBUF;fourth,starttogatherimageandplacetheminthebufferzoneusingVIDIOC_STREAMON;fifth,VIDIOC_STREAMOFFiscalledtostopcollectingimage.Atthispoint,thedriverwillautomaticallyremoveallthebuffersfromimageacquisitionsequence,andclearallthememorymappingsbymunmap().Note:beforereadingbufferzone,VIDIOC_DQBUGshouldbecalledatfirsttonoticethatthebufferisunavailable,oritwillleadtoinconsistenceofrelatingimages.IV.RESULTANDPROSPECTThefollowingfigure3showsthehardwareplatformandtheresultofexperiment.Thecollectingpicture,whosesizeis400480,displayedontheLCDclearly.BasedonLinuxplatform,theimageiscollectedbyCMOScameraandpreviewedsuccessfullyusingimageacquisitionapplicationsonthebaseofV4L2.Duringtheoperationinactual,thepicturesdisplaywellandhavegoodquality.Theadvantagesofthesystemaregoodscalabilityandextensiveuse,andsomereferencevalueissuppliedforrelevantareasofvideoimageacquisition.Forthenextstep,itishopedthatthepicturegatheredwillbeencodedandtransmittedtoPCtorealizevideosurveillance.ACKNOWLEDGEMENTWewishtothankProfessorHuangZhiweiofDepartmentofElectricalEngineeringatUniversityofSouthChinaforhisencouragementandsupportonthispaper.Besides,wealsoappreciateseniorsatthelaboratoryfortheirsuggestionandproofreadingspellingforthepaper.REFERENCE1SongLili.ImageCaptureandPretreatmentDeviceDesignBasedonEmbeddedSystem.JournalofHarbinUniversityofScienceandTechnology,2010,15(3):23-262SamSungElectronics.S3C6410microprocessorusersmanual.www.S,20093OminiVisionTechnologiesInc.OV9650ColorCMOSSXGA(1.3MegaPixel)CamerachipTMImplementationG,20044DirksB.VideoforLinuxTwoAPISpecificationDraft0.24./spec/book1.htm,20085LiuYakun,ChenXiaodong.ThestudyofvideoacquisitionandcompressioncodecsystembasedonDaVinci:2010InternationalConferenceonInformationNetworkingandAutomation,2010(2)V2-317V2-3216LiuYonggui,FanLimin,HuFuqiao.StudyBasedonImplementMethodofVideoCaptureSystem.ModernElectronicTechique.2008(7):61-647ZhengJiehang.TheDesignofNetworkVideoMonitoringBasedonARM.MsDThesis.Wuhan:WuhanUniversityofTechnology,2010中文翻译基于V4L2嵌入式图像采集的实现摘要：考虑到基于V4L2（VideoforLinuxTwo）设计的Linux系统嵌入式图像采集应用程序较少这一现状，设计了基于V4L2的嵌入式图像采集应用程序。采用S3C6410作为硬件开发平台和Linux作为软件开发平台。在安装了CMOS摄像机OV650驱动以及其他相关驱动程序后，编译了基于V4L2的图像采集应用程序。为了减少CPU利用率，基于V4L2的图像采集应用程序采用双帧缓冲实现图像的采集和预览。在图片大小为400480和帧频为15帧/秒的条件下，采集到的图片很清晰而且质量很好。实验结果表明，该系统具有良好的可扩展性,而且它为嵌入式便携式设备提供了一个新的想法。关键字：V4L2，OV9650，图像采集，S3C6410，双帧缓冲器一、介绍视频图像处理技术广泛应用于电子技术和计算机技术。从照相机或其他图像采集设备上得到的图像以数字的形式转换，得到数字图像信息；然后，经过计算机的编码，过滤，增强，恢复，压缩，储存以及其他过程；最终呈现出视觉图像或其他结果。随着嵌入式技术的发展，现在的图像采集技术实现了从模拟到数字的转换。由于嵌入式图像采集技术具有优越的性能，受到了越来越多的人们的关注。对于图像处理系统来说，嵌入式图像微处理器是最好的选择，因为其具有高性能，高成本效益以及低功耗等优点。三星S3C6410具有ARM11核并且采用了64/32位内部总线架构。它还包括了许多强大的硬件加速器，比如视频处理，音频处理等等。一个完整的MFC(多格式编解码器)支持MPEG4/H.263/H.264的编码和解码以及VC1的解码，所以对于图像的处理是非常的方便。S3C6410被用作嵌入式硬件平台的核心，本文设计了基于V4L2的图像采集应用程序。把支持OV9650和V4L2的Linux操作系统内核和移植添加到硬件平台，最终图像被搜集并显示在液晶屏上。二、硬件系统结构整个图像采集系统主要包括CMOS摄像机OV9650，内核控制器S3C6410，液晶屏和其他外围设备。系统的硬件结构如图1：A处理器：三星S3C6410是基于ARM1176JZF-S核心，并且它的主要电源频率是533MHz/667MHz，包括媒体的SIMD(单指令多数据)处理。它适用于特殊的方法来使它适合视频处理。B.摄像机：这里使用的是OmniVision的OV9650摄像机。它集成了一个10位a/D转换器,对应于数据输出接口D0:9。通过内部DSP处理，输出图像数据格式可以10位原始RGB或8位RGB/YcbCr。使用的CAMIF支持ITU-BT601/656YCbCr8位模式。因此，OV9650的D9:2应该被链接到S3C6410中的CAMIF数据接口。C.外围设备：这部分主要包括液晶屏和触摸屏接口电路，JTAG调试电路，复位电路，电源电路，RS232串行接口电路等等。因为OV9650寄存器控制协议SCCB(一系列摄像机控制总线)可以兼容I2C，所以我们可以使用I2C作为后期平台模拟SCCB安装OV9650内部寄存器。三、软件设计有两个步骤来实现图像的采集:第一，在Linux内核底部加载相关驱动，第二，为了得到图像数据，在顶部编译图像采集应用程序，图像最终会显示在液晶屏上。A装载相关驱动在编译应用程序之前，应该先装载相关的驱动。因为，被采用的内核中包括驱动，所以仅仅添加相关的配置到内核中。1）内核配置在内核清单中，内核配置使用命令“makemenuconfig”：选择设备驱动程序多媒体设备Linux操作系统上视频软件，装载video4linux模块为视频设备提供编译接口；选择设备驱动程序多媒体设备三星相机接口驱动器外部相机OVTOV965x增加对CMOS图像传感器的支持到内核。保存配置并推出。2）内核图像的生成使用下面的指令：执行数据保护：读取配置文件并创造和配制相一致的相依树状结构来确定编译选项。制作图像：生成可以支持V4L2和OV9650的内核图像。在内核启动时，如果有如下的信息，表明OV9650驱动已成功加载：CAMOV9650initregstart.CAMOV9650initregend.在系统开始之后，它可以自动识别OV9650。系统调用可以用来获取设备信息来帮助应用程序的开发，CMOS图像传感器驱动提供的I/O操作界面可以实现相关的操作。B.图像采集应用程序V4L是由Linux提供的对于视频设备应用程序的一系列接口功能，图像数据可以使用API功能从视频设备上读出。1）V4L2驱动结构V4L有两个缺点：当驱动程序在被设计的时候，你不可以同时打开多个设备；而且，V4L2的API不能很好的支持有编译功能的设备，在这种情况下，BillDirks设计了一系列的API和数据结构，被叫做V4L2。和V4L相比，提高了可拓展性和灵活性，越来越多的视频设备包含在其管理之下。尽管V4L2是V4L的更新版本,但它们不兼容。V4L2被应用于Linux2.6内核驱动协议中。作为Linux系统中一个标准的视频捕捉装置：为了驱动的开发，V4L2采用了分层的方法来提供明确而始终如一的接口：应用程序在顶层，V4L2在中间层，而硬件装置实际在底层。V4L2有双重驱动体系结构。上面是videodev模块。当videodev被初始化，就指示自己为主号码为81的字符设备，并且它们自己的字符驱动数字功能被同时注册。较低的是V4L2驱动，事实上，videodev可以被叫做通过V4L2驱动数字功能的Videodev。在V4L2驱动被初始化后，包括V4L2驱动元函数，小数字和其他相关信息在内的结构都被传递给videodev，结果是，被手动操作的装置被注册到videodev。当V4L2被初始化时，首先要计算被处理的系统装置；然后，V4L2-装置被加到每一个设备中，并且指向V4L2-装置的指示器被传到V4L2-寄存器-装置中，这被称作为初始化设备而初始化V4L2-装置的功能。在被称作V4L2-寄存器-装置之前，信息的名字，类型，小数字和开放功能必须被