已阅读5页,还剩1页未读, 继续免费阅读
版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
TheNinthInternationalConferenceonElectronicMeasurement&InstrumentsICEMI2009ANearInfraredImagingDetectionSystemBasedonDavinciPlatformLiHua,ZhangShi-chao,HanChao,ZhengMing,MengXiao-fengLiHua:Professor,DepartmentofPhysics,BeihangUniversity,Beijing,ChinaEmail:ZhangShi-chao:DepartmentofPhysics,BeihangUniversity,Beijing,ChinaEmail:AbstractInfraredimagingdetectiontechnologyisahigh-tech,involvingofoptics,electronics,machineryandcomputerscienceinmanyfields.Thetechnologynowplaysacrucialroleindetectionofearlywarning,identificationtracking,publicsecurity,firealarmandapplicationsinmanyotherfields.Thispapermainlyexpandsonanear-infraredimagingdetectionsystembasedonDavinciDigitalSignalProcessor(DSP)platform.Andthisarticledescribestheschematic,hardwarestructureandimagedataprocessingalgorithmofthenear-infraredimagingdetectionsystemindetail.Firstly,infraredradiationfromthetrackedtargetisfocusedontheCMOSnearinfraredimagesensorthroughaninfraredopticallens.TheimagesensorcontrolledbyaComplexProgrammableLogicDevice(CPLD)producesthecorrespondingimagedata,andthentheimagedataisscaledbytheCPLDtofitthefollowingdataprocessing.TheCPLDalsoplaysakeyroleinvideobufferingandnoisereduction.TheDavinciDSP-SOCwillthenacceptthedataflowafterward,andcalculatethelocationofthetargetwiththecorrespondingimageprocessingalgorithmstocontroltheservosofhorizontalandvertical,lockingandtrackingtheinfraredtargetinrealtime.TheARM9coreintheDSP-SOCisoperatedwithLinux,whichisfreeandopensource.TheusercanhumanlyprogramthecoreDSPalgorithmsandtrackingalgorithmswithaPCtogetamuchbettereffectoftracking.Withatouch-screenandTFTLCD,thesystemismuchmoreconvenientanduser-friendly.Duetothehighsensitivityandresolution,thenearinfraredimagingdetectionsystemcanbeusedatnightorinverylowvisibility,havinganexcellenttargetrecognitioncapabilityandprecision-detectioncapability.KeywordsInfraredimagedetection,Infraredimageprocessing,CMOS,DSP-SOC,ARM9.I.INTRODUCTIONInfrareddetectiontechnologyisahigh-techusingofinfrareddetectorstocaptureinfraredradiationtosearchandtrackthetarget.Itisfeaturedbyhighprecision,freefromtheeffectsofradiofrequencyinterferenceandcanbeoperateddayandnight.Thetechnologyisusuallyusedinfirealarmandpublicsafety.Anditcanberoughlydividedintonon-imagingandimagingdetectiontechnology.Non-imaginginfrareddetectiontechnologymakesuseofoneorseveralinfrareddetectorstocaptureandtrackthetarget,forwhichgivingoutinfraredradiationduetothermalmovementoftheatomsandmolecules.Becauseofthestructure,itcanbeeasilyinterferedbythenormalworkofclouds,foganddustimpact.Infraredimagingdetectionisusuallymakinguseofinfrareddetectorarraytodetecttheinfraredradiationgivenoutbythetarget.ThedetectorarrayisusuallyaCCDorCMOSimagesensorandtheimagequalityissimilartotelevision.Butitcanworkatnightandinverylowvisibility,whichthetelevisiondetectionsystemcanbedifficulttoworkin.Infraredimagingdetectiontechnologyhasnowbecomeamajordirectionofthefutureuse.Therearemanywaystoachieveinfraredimaging,mainlyusingthefollowingtwoways:(1)Multiplescanninginfraredimagingdetectorlineararrays;(2)Multiplenon-scanninginfraredimagingdetectorplanararrays.Theinfraredimagingdetectorhasbeendevelopedquicklysincethe1970s,fromlineararraytotheplanararray,andfromnear-infraredtofarinfrared.Thenumberofelementsofinfrareddetectorplanearrayisincreasing.Infraredimagingdetectionsystemhasagoodsensitivityandhighspatialresolution,awiderangeofdynamictrackingaswellastheeffectiverange,comparedtothenon-imaginginfrareddetection.Infraredimagingdetectionsystemhasbetterabilitytocompletetargetidentificationandprecisiondetection.Foraninfraredimagingdetectionsystem,theapplicationofimageprocessingandtargetrecognitionisakeyfeaturedifferencefromnon-imaginginfrareddetectionsystem.Thesensitivityofthedetectorarray,theperformanceofimageprocessingandtargetrecognitionalgorithm,theperformanceofdetectionalgorithmandthemechanicalmobilityalldeterminetheperformanceofthedetectionsystem.ThispaperwilldeepintothesearchofanearinfraredimagingdetectionsystembasedonDavinciDSPplatform,whichissuitableforvideoandimageprocessingduetoitsstructure.TheplatformhasanARM9coretransplantedwithLinuxoperatingsystemandaC64x+DSPcoretocompleteimagingalgorithmsinrealtime.Thepaperwillbestructuredasfollows;InSectionwewillexposetheframeworkofthesystem;InSectionwewilldeepintothestudyofhardwaredesignoftheimageprocessingandrecognition;InSectionwewilldescribeourapproachofsoftwarestructure,especiallytheLinuxoperatingsystemmigration;Insectionwewilldeepintothestudyoftheimageprocessingand4-154_978-1-4244-3864-8/09/$25.002009IEEETheNinthInternationalConferenceonElectronicMeasurement&InstrumentsICEMI2009recognitionalgorithmsandmethodssuitablefortheinfraredimagingdetectionsystem.II.SYSTEMFRAMEWORKThesystemcanberoughlydividedintothreeparts,thehardwarepart,thesoftwarepartandmechanicalpart.Allpartsaboveareanorganicwholeforthecloserelationmutually.Andtheblockdiagramofthesystemisshowedinfigure1below.Fig.1.BlockdiagramofthenearinfraredimagedetectionsystembaseonDavinciplatformAsshowninfigure1,theinfraredradiationfromthetrackedtargetisfocusedontheCMOSnearinfraredimagesensorthroughaninfraredopticallens.WithaCPLDthecorrespondingimagedataflowisscaledandbuffered.TheDavinciDSP-SOCwillacceptthedataflowafterward,andthencalculatethelocationofthetargetwiththecorrespondingimageprocessingalgorithms.Sincehasgotthelocationinformationofthetarget,theDSPthencontrolstheservosofhorizontalandvertical,lockingandtrackingtheinfraredtargetinrealtime.TheARM9coreintheDSP-SOCisoperatedwithLinux,whichisfreeandopensource.TheusercanhumanlyprogramthecoreDSPalgorithmsandtrackingalgorithmswithaPC.Withatouch-screenandTFTLCD,thesystemismuchmoreconvenientanduser-friendly.Duetothehighsensitivityandresolution,thenearinfraredimagingdetectionsystemcanbeusedatnightandinverylowvisibility,havinganexcellenttargetrecognitioncapabilityandprecision-detectioncapability.III.THEHARDWAREDESIGNTocapturetheinfraredimagesignalinverylowvisiblecircumstance,imagesensorwithhighsensitivityatnearinfraredbandshallbeapplied.ThemostpopularlyusedsensorsareCCDandCMOSimagesensor.Consideringthefeaturedifferencesofthebothintable1,CMOSimagesensorischoseninthedesignforitshighsensitivityaroundnearinfraredband,lowcostandlowpowerconsumption.Table1.FeaturedifferencesbetweenCCDandCMOSimagesensor.ParameterSensitivityatinfraredbandCostNoisePowerConsumptionCCDModerateHighLowHighCMOSHighLowHighLowThekeypartofthehardwareistheDSP-SOC,includinganARMcoreandaDSPcoreonthechip.TheARMcoreisusedasthemastertohandlecomplexsystemworkandDSPcoreastheslaver,handlingvideoprocessingandrecognitiontask.Togiveamuchmoreclearviewofthehardwarestructure,theblockdiagraminfigure2willbrieflyandclearlypresentsthevideoprocessingsubsystem(VPSS)onthechip.Fig.2.BlockdiagramofthevideoprocessingsubsystemThevideoprocessingsubsystemprovidesaninputinterface(videoprocessingfrontend,VPFE)forexternalimagingperipheralssuchasimagesensors,videodecoders,etc.;anoutputinterface(videoprocessingbackend,VPBE)fordisplaydevices,suchasanalogSDTVdisplays,digitalLCDpanels,HDTVvideoencoders,etc.Inadditiontotheseperipherals,thereisasetofcommonbuffermemoryandDMAcontroltoensureefficientuseoftheDDR2burstbandwidth.Thesharedbufferlogic/memoryisauniqueblockthatistailoredforseamlesslyintegratingtheVPSSintoanimage/videoprocessingsystem.ItisimperativethattheVPSSutilizeDDR2bandwidthefficientlyduetobothitslargebandwidthrequirementsandthereal-timerequirementsoftheVPSSmodules.BecauseitispossibletoconfiguretheVPSSmodulesinsuchawaythatDDR2bandwidthisexceeded,asetofuser-accessibleregistersisprovidedtomonitoroverflowsorfailuresindatatransfers.ThereforetheDSPissuitablefortheimagedetectiontask.AstheDDR2offershigh-bandwidthforthesystemoperation,thequalityofthePCBdesignofDDR2willaffectsthestabilityandrobustnessofthewholesystem.AsthehighestfrequencyonboardisnearlyGigahertz,themostchallengingworkinthedesignishowtoresolveElectroMagneticCompatibilityofthewholesystem.4-155TheNinthInternationalConferenceonElectronicMeasurement&InstrumentsICEMI2009Asrequired,thePCBshouldbeatleastappliedwithsixlayerstoresolveEMC,andtheminimumPCBstackuprequiredisshowninTable2.Eachsignallayerhasagroundplanetorefertoformmicrostripsandstriplines.Themultilayerdesignalsobenefitstothesignalintegrityandpowerintegrity.Table2.MinimumPCBstackuprequired.TheroutingofDDR2willbediscussedindetailinthefollowingparagraphs.TheCKandADDR_CTRLnetclassiscompletelysourcedbytheDSPtotheDDR2devices.EachnetisabalancedTroute,seeFigure3.ThelengthofsegmentAshouldbemaximizedandtheoveralllengthfromAtoBorAtoCshouldbeminimized.Ideally,thePCBdelayoftheCKnetclassisidenticaltothedelayfortheADDR_CTRLnetclass.AllnetsintheCKandADDR_CTRLnetclassesarematchedinlengthtoeachotherwithin100mils.AndthenetsintheCKnetclassarelaidoutasadifferentialpairtoachievehighreliabilityandnoiseimmunity.OthertracesshouldbekeptawayfromtheCKnetclasstracesbyatleast4wcenter-to-centerspacing(recallthatw=minimumtracewidth/space).TraceswithintheADDR_CTRLnetclassshouldbespacedatleast3wcenter-to-centerfromeachother.Tracesofothernetclassesshouldbekept4wawayfromtheADDR_CTRLnetclass.Fig.3.TopologyRequirementsforADDR_CTRLandClockNetClassesTheeightnetclassesthatmakeupthefourDQSBsandfourDQBbyteshavethesameroutingrules.NotethattheskewmatchingisrequiredbetweentheDQBnnetclassanditsassociatedDQSBnnetclass.ThesenetclassesaresourcedbytheDSPdeviceduringwritesandaresourcedbytheDDR2devicesduringreads.Ideally,thePCBdelayoftheDQSBnnetclassisidenticaltothedelayfortheDQBnnetclass.AllnetsintheDQSBnandDQBnnetclassshouldbematchedinlengthtoeachotherwithin100mils.ThelongesttracepermissibleisequaltothelongestManhattandistanceoftheDQSBnandDQBnnetclasses.OthertracesshouldbekeptawayfromtheDQSBnnetclasstracesbyatleast4wcenter-to-centerspacing.TraceswithintheDQBnnetclassesshouldbespacedatleast3wcenter-to-centerfromeachother.Tracesofothernetclassesshouldbekept4wawayfromtheDQBnnetclass.Togiveaclearviewofthehardwarerouting,thePCBoftheDSPtoDDR2isshownindetailinfigure4below.Fig.4.PCBoftheDSPinterfaceswithDDR2BeforethePCBwasmachining,simulationwassuccessfullycarriedoutwith133MHzclockrateand533MHzdataratetoensurethatthesystemwouldworkproperly,andthesimulationwaveofaddressbusanddatabusisshowninfigure5and6respectively.Fig.5.SimulationofDDR2addressesbusandclockbusat133MHz.Fig.6.SimulationofDDR2databusat533MHz4-156TheNinthInternationalConferenceonElectronicMeasurement&InstrumentsICEMI2009IV.THESOFTWAREDESIGNLinuxisaUnix-likeoperatingsystem.Afterdecadesofdevelopment,theLinuxoperatingsystembecomesmoreandmorematureandpopular.Featuredasopensource,theLinuxOSisfreeandflexibletouse.Furthermore,Linuxissuitableforworkstation,server,personalcomputerandembeddedsystem,foritsupportsmultiplearchitectures,CISCandRISC,suchasX86,ARM,MIPS,andsoon.Anembeddedapplicationsisusuallytransplantedwithanoperatingsystem,andLinuxisamostpopularonearoundtheworld,foritisopensourceandwithhighreliability.WewillmakefullexploitoftheLinuxOSanditstransplantationinourdesign.Firstofall,thekernelistheheartoftheLinuxoperatingsystem,whichoffersmanagementofprocess,managementofmemory,filesystem,devicecontrolandnetwork.Thefunctionblockisshowninfigure7.Fig.7.FunctionblockofthekernelofLinuxTheprocessmanagementmoduleisakeyblockoftheLinuxkernel,takingchargeofcreatingandterminatingprocess.Communicationthroughsignal,pipelineorcommunicationprimitivesamongdifferentprocessisabasicfunctionforthewholesystem,whichisaccomplishedbythekernel.Thememorymanagementmoduleistoensurealltheprocessescansafelysharethesystemmemory.Andwhatsmore,itshouldalsosupportvirtualmemorymodetohighlyimprovetheefficiencyofmemoryusage.Offeringacommoninterfaceforperipherals,thefilesystemmoduleisusedtosupportstorageanddriversforperipheralsandthevirtualfilesystemhidesdetailsofdifferenthardware.BootLoaderisthecoderunimmediatelyafterpoweron.WiththeBootLoader,hardwaredevicescanbeinitialized,andmassagescanbecreatedanddeliveredtothekernelthroughthecorrespondingmechanisms.AndthentheBootLoaderbringsthesystemtoapropermodeandloadthekernel.Figure8showsthebasicflowchartoftheBootLoader.TheBootLoaderiscloselydependedonthehardwareenvironment.Andbesidesofarchitecture,theBootLoaderalsodependsontheboardconfigurations,especiallyfortheembeddedapplications.Fig.8.FlowchartoftheBootLoaderAftertheBootLoaderinitializethewholesystemandthenreleasetheauthoritytotheLinuxkernel,thekernelthenworks.Thedevicedriveristheroutineworkingbetweenthehardwareandapplicationsinthekernelspace.Naturedasaconverterbetweenlogicdevicesandphysicaldevices,itstaskistoinitializeandmanipulatethecorrespondingI/Odevices.Thedevicedrivermasksdetailsofthehardwarefortheapplications.Withthedevicedriver,theapplicationscancontrolhardwaredeviceaseasyasmanipulatecommonfiles.Theapplicationsoftwareoftheinfraredimagedetectionsystemisdesignedtoaccomplishtheimageprocessingandrecognitiontaskswiththecorrespondingalgorithms.Thealgorithmswillbedetailedinthefollowingsection.V.IMAGINGPROCESSINGANDRECOGNITIONALGORITHMSANDMETHODSTheimageprocessingandrecognitionprogramiscompletedontheapplicationlayerofLinux.Andtheflowchartoftheapplicationisshowninfigure9.Besidesotherlayersdesign,wewillfocusontheimageprocessingandrecognitionalgorithmsandmethodsusedintheapplicationlayeraccordingtotheflowchart.Theinfraredimageisusuallymonochromeorpseudo-colouredandwithlowcontrast,sometimesmightbeunclearordifficulttorecognize.Forexample,itmightbenoteasytofigureoutthebirdhiddeninlushfoliageinapicturetakenatnightorinverylowvisibility.Imageenhancementisintroducedtoenlargethedetailsingrayscaleorspatialtoobtainaclearimage.Therearesomemethodsusedinourdesign,suchasgrayscaleenhancement,contrastenhancementandhistogramadjustment.Contrastenhancementmagnifiesntimesthegrayvalueoftheentirepixels,thatis;g11g12g11g12g,xynfxyg32g1174-157TheNinthInternationalConferenceonElectronicMeasurement&InstrumentsICEMI2009Thegrayvalueintheoriginalimageisg11g12,fxyandthegrayvalueinthecontrastenhancedimageisg11g12,gxy,whichisntimesthegrayvalueintheoriginalimage.Histogramadjustmentisamethodtocompressthegrayvaluewithfewerpixels,andextendthegrayvaluewithmorepixels.Aftercontrastenhancementandhistogramadjustment,detailsintheimagecanbeexposed.Andthecorrespondinghistogramisflatteredaftertheprocess.Theoriginalimageoftendistributedwithrandomnoise,makingtheimagedeteriorated.Inordertogetridofsuchnoise,imagesmoothingormedianfilterareused.Withtherandomnoiseoftheoriginalimagebeingremoved,sharpdetailsarealsolostduringtheprocess.Fig.9.FlowchartoftheapplicationAftertheimageprocessing,imagerecognitionisfinallyusedtofulfilltheinfraredimagingdetectiontask.Edgedetectionandextractionbasedondifferentialisapplied.Duetothedramaticchangesingrayvalueofthebrink,itisclearthatdifferentialfunctioncanwellextractthechanges.Supposeg11g12,xypresentsthelocationofapixel,andg11g12,fxypresentsthegrayvalueofit.Withavectorg11g12g11g12,xyGxyffg32,theedgecanbeeasilyfiguredoutbythegradient.xfandyfarethedifferentialalonghorizontalandvertical,calculatedwiththefollowingequations.g11g12g11g12g11g12g11g12g961,1,xyffxyfxyffxyfxyg32g14g16g32g14g16Sotheintensityoftheedgeis22xyffg14,andthedirectionisalongthevectorg11g12,xyff.Anthermethodusuallyusedisedgedetectionandextractionbasedontemplatematching.Andtemplatematchingisresponsibleforresearchingtheconsistencyofimageandthecorrespondingtemplate.Patternrecognitionisappliedtoclassifypatternsfromeachothertodeterminewhethertheobjectisthetargetornot.Andthefollowingmethodsarethechoices;Templatematching;statisticalclassification;andartificialneuralnetwork.Templatematchingistheearliestandeasiestwayforimagerecognition,whichissomewhatoneofstatisticalrecognition.Matchingisageneraloperationdenotedbycorrelationcoefficient,usedtodefinethesimilaritybetweenthetemplatesandthesamples.Whenusingtemplatematching,templ
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 2026年西安电力机械制造公司机电学院单招(计算机)考试备考题库及答案1套
- 2026年平顶山文化艺术职业学院单招职业适应性测试题库附答案详解
- 2026年黄河交通学院单招综合素质考试题库附答案
- 2026年辽宁省朝阳市单招职业倾向性考试题库附答案
- 2025河北雄安容西青禾幼儿园、临泉幼儿园教师见习岗招聘(公共基础知识)综合能力测试题附答案解析
- 2026年江汉艺术职业学院单招职业适应性测试题库及参考答案详解1套
- 2026年西昌民族幼儿师范高等专科学校单招职业适应性考试模拟测试卷附答案
- 2026年黑龙江护理高等专科学校单招职业技能考试题库及参考答案详解一套
- 2026年江苏护理职业学院单招职业倾向性测试题库附答案详解
- 2026年苏州高博软件技术职业学院单招职业技能测试题库含答案详解
- 2025年低压电工证题库(详细版)
- 2025年计算机应用基础期末考试试题
- 破产管理人培训课件大纲
- 冷菜承包协议书范本
- 2025至2030中国正畸矫治器行业项目调研及市场前景预测评估报告
- 气血不足的课件
- 乌兰木伦村“煤海探秘”工业旅游开发方案探索
- 2025地球小博士知识竞赛试题附答案
- 2025广东省南粤交通投资建设有限公司招聘笔试历年备考题库附带答案详解试卷2套
- 2025年公营养师(四级)试题及答案
- 2025年国考《行测》全真模拟试卷一及答案
评论
0/150
提交评论