會(huì)計(jì)學(xué)1ComputerVision計(jì)算機(jī)視覺(jué)英文實(shí)用Computervisionresearch

Attheendofthe1950s,computershad

becameanimportanttoolforhumanbeings

torealizethehumanintelligenceandhumanperception,withthehelpofcomputer,Forthefirsttime,humanbeingscanrealizetheextensionofmentalandperceptualabilityaswellastheextensionofphysicalstrengthwiththeaidofthemachine.thesimulationofhumanvisualperceptionleadstothegenerationofcomputervision.Specifically,computervisionwillachievethreebasicpurpose.1Thefirstpurposeisto

calculatethedistancefromtheobservationpointtotheobjectaccordingtooneormore2Dimages;第1頁(yè)/共23頁(yè)

2Thesecondpurposeisto

calculatetheMotionparametersfromtheobservationpointtotheobjectaccordingtooneormore2Dimages;3Thethirdpurposeisto

calculatethephysicalcharacteristicsfromtheobservationpointtotheobjectaccordingtooneormore2Dimages;Therearetwoultimategoalthatcomputervisionwillrealize,theycanbesimplydescribedasfollow:

oneistorealizetheunderstandingoftheworld,inotherwords,

itcanrealizesomefunctionsofthehumanvisualsystem;theotheristorebuildthethreedimensionalobject

accordingtothetwo-dimensionalprojectionimages.第2頁(yè)/共23頁(yè)Historyofcomputervision1950s:inthisperiod,statisticalpatternrecognitionismostappliedincomputervision,itmainlyfocuseontheanalysisandidentificationoftwo-dimensionalimage,suchas:opticalcharacterrecognition,thesurfaceoftheworkpiece,theanalysisandinterpretationoftheaerialimage.1960s:Roberts(1965)useacomputerprogramtoextractethethree-dimensionalstructureofapolyhedronfromthedigitalimages.healsodescribetheshapeandthespatialrelationshipofobjects.Robertscreatedanewsituationfortheresearchofcomputervision.hispurposeofresearchbegantobecometheunderstandingthethree-dimensionalscene.thecreativeresearchfortheblocksworldgivepeoplegreatinspiration.manypeoplebelievethatoncethethree-dimensionalworldmadeupofthewhitebuildingblocktoycouldbeunderstandable.sitalsocanbeextendedtounderstandmorecomplex3Dscene.第3頁(yè)/共23頁(yè)

theresearchscopeoftheblockworldisextensive.itincludesthefeatureextractionofedgeandcorner,thegeometricfactoranalysisoflineandplane.Italsobuildavarietyofdatastructuresandinferencerules.Inthemiddleof70's,artificialintelligence(AI)laboratoryofMassachusettsInstituteofTechnology(MIT)officiallyopenedthe"machinevision"course.Atthesametime,AIMITlaboratoryhasattractedmanyfamousscholarsfromtheworldtoparticipateintheresearchofmachinevision,whichincludedthetheoryofmachinevision

,algorithmanddesignofsystem.ProfessorDavidMarrin1973wasinvitedtoleadaresearchteamwhichisconsistedofdoctoralstudentsintheMITAILab.in1977,DavidMarrputforwardanewcomputervisiontheorywhichisdifferentfromthepasttheorywhichisbuildontheanalysismethodof"blocksworld".inthe1980s,histheoryhadbecomeaveryimportanttheoreticalframeworkinthefieldofmachinevision.第4頁(yè)/共23頁(yè)

Changesinthenameofapublicationcanreflecttheprogressofcomputervision:1972,CGIP(computergraphicsandimageprocessing)publication;In1983,itsnameisCVGIP(computervision,graphicsandimageprocessing);In1991,thepublicationisdividedintotwoversions,whichareCVGIP-GMIP(graphicsmodelandimageprocessing)andCVGIP-IU(imageunderstanding);第5頁(yè)/共23頁(yè)

In1995,thepublicationbecometwomagazine.theformernamedGMIP(graphicsmodelandimageprocessing),thelatternamedCVIU(computervisionandimageunderstanding).Theupsurgeofresearchoncomputervisionbeganin1980s,andinthemiddleof80's,thecomputervisionhasdevelopedrapidly,newconcepts,newmethodsandnewtheorieshaveemerged.Sofar,computervisionisstillaveryactiveresearcharea.第6頁(yè)/共23頁(yè)AdvancesincomputervisionresearchAfternearly50years,researchincomputervisionhasmadeconsiderableachievements.ThismakesthefieldtobecomethemostactivepartoftheAIresearchinthepast40years.Itsmainstreamresearchisdividedintothreestages:Stage1:Researchonthevisualbasicmethod,whichtakethemodelworldasthemainobject;Stage2:Researchonvisualmodel,whichisbasedonthecomputationaltheory;Stage3:Researchonthecomputervisionmethodwiththegoalofapplication;第7頁(yè)/共23頁(yè)stage1

In1965,Robertsuseacomputerprogramtoextractethethree-dimensionalstructureofapolyhedronfromthedigitalimages.healsodescribetheshapeandthespatialrelationshipofobjects.IntheworkofRoberts,therelationshipbetween3Dobjectsand2Dimagingisintroduced,andsomesimplemethodsof

featureextractionareintroduced.Later,peoplebegantomakesadeepstudyofblockworldandestablishavarietyofdatastructuresandinferencerules.Thescopeofthestudyisabroad,whichincludethefeatureextractionofedgeandcorner

,geometricelementsanalysisoflineandflatsurface,surface.第8頁(yè)/共23頁(yè)

,Thestudyofblockworldreflectssomeofthecharacteristicsofearlyvisualresearch,thatisstartingfromasimplifiedworld.Theworkinthisperiodhasplayedanimportantroleinthedevelopmentofcomputervision,butitisdifficulttoworkforaslightlycomplexscene.Researchonthisstageisbasedontheanalysisofthreedimensionalspatialrelations,moerover,analysisofthree-dimensionalrelationshipisonlyrelyonsimpleconstraintrelations.第9頁(yè)/共23頁(yè)stage2Beginningin1970s,thestudyofcomputervisionbegintocomeintoamorerationalstage.1peoplebegintostudytheimagingmethodsfromthepointofviewofphysiology,opticsandprojectivegeometry.2Reasearchismainlyconcentratedintherecoveryofthevariousintrinsiccharacteristics,includingtherecoveryofthree-dimensionalshape,motion,lightsource,etc..Inthisstage,someresearchers,representedbyMarr,putforwardageneralvisualprocessingmodel,.第10頁(yè)/共23頁(yè)stage3Afterenteringthe1990s,withthereflectionontheresearchofartificialintelligence,peoplearestartingtothinkaboutwhetherthepastmehtodsiscorrect,whetherthetheoryhaspracticalsignificanceintheareaofcomputervision.researchoncomputervisionwastransferredfromthesimpletheoreticalframeworktothecombinationofavarietyofapplications.Someoftheimportantcharacteristicsofbiologicalvisioninthepasthavenotbeenpaidenoughattention.Therefore,intheearly90's,somescientisthasputforwardnewmethodsandnewideasofactivevisionandqualitativevision.Byusingactivegeometricmethod,somecomplexcomputerscanbesimplified.第11頁(yè)/共23頁(yè)TheapplicationofcomputervisionIndustrialvision:peoplegettheinternalimageoftheobjectbyX-rayorultrasonicdetectiontoachievethesurfacedefectinspection,internalnon-destructivetesting,etc.;Throughcamerasandrelatedvisualinformationprocessingsystem,peoplecouldchoosethedifferenttransmissionbeltaccordtothedifferentobjects，thenseparatetheobjectfromthetransmissionbeltandidentifyandlocateit.第12頁(yè)/共23頁(yè)

：Humancomputerinteraction:peoplecanletthecomputerusethedemandsofthehumangestures,lipmovements(lipreading),physicalexercise(gait),facialexpressiontounderstandpeople'sdesire,whichisinlinewiththehabitofhumaninteraction,italsocanincreasetheconvenience.Suchasfacerecognition,intelligentagents,etc..人第13頁(yè)/共23頁(yè)Visualnavigation:Peoplecanusetwoormorecamerastogetasetofimagestorecover3Dsceneinformation,andusethese3Dsceneinformationtorecognizetarget,toidentifytheroad,identifyobstacles.

Stereoimagesacquiredbyastereovisionsystemcanbeusedtoreconstruct3Dsceneinformation.第14頁(yè)/共23頁(yè)Visualnavigation：Thecombinationofstereoscopicimagesandmotioninformationcanbeascenegraphtomeetthespecifictask.Suchascruisemissileguidance,unmannedaircraftflying,automaticmovingvehicles,mobilerobots,etc.,notonlytoavoidtheriskofhumaninvolvement,butalsotoimprovetheaccuracyandspeed.

CourageRoverRocky7Marsrover第15頁(yè)/共23頁(yè)

Securityidentification,monitoringandtracking:：Parkingareamonitoring,vehicleidentification.Machinecanidentifyaspecificpersonaccordtothefeaturesoftheface,theeyeground,thefingerprint;Atpresent,peoplearestudyingakindoffacialmotionparametersinordertoanalyzethehumanexpressionandinneractivities

bytheextractionanddescriptionoftheparameters.第16頁(yè)/共23頁(yè)

Medicalimageanalysis：Therearetwoaspectintheapplicationofcomputervisioninmedicalimagediagnosis:First,theimageenhancement,labeling,dyeingandotherprocessingcanhelpdoctorstodiagnosethedisease,andtoassistdoctorstomeasureandcompare

theareaofinterest;second,itcanusetheexpertknowledgesystemtoanalyseandexplanationanimage(oraseriesofimagesinaperiodoftime).

第17頁(yè)/共23頁(yè)Contentsofcomputervisionresearch1）(inputdevice)includingimagingequipmentanddigitalequipment.ImagingdeviceisadevicewhichcandetectthesceneorobjectbytheopticalcameraorXrays,andgetthe2Dor3Ddigitalimageofthesceneorobject.2）(lowlevel)Mainlytodealwiththeinputimage.Thisprocessusesalargenumberofimageprocessingtechniquesandalgorithms,suchasimagefiltering,imageenhancement,edgedetection,texturedetection,motiondetection,inordertoextractthebasicfeaturesofthescene.第18頁(yè)/共23頁(yè)

3）（middlelevel）Themaintaskistorestorethedepthofthescene,thedirectionofcontactnormal,contourandother2.5dimensionalinformationoftherelevantscene,themethodtoachieveitincludestereovision、rangefinder.4）（highlevel）

themaintaskhasthreeaspect,oneistoreconstructthecomplete3Dgraph,thesecondistobuildthe3Ddescriptionofobject,thethirdistorecognizeanddeterminethepositionandorientationoftheobject.theseprocessisbasedontheoriginalinputimage,thebasiccharacteristicsoftheimage.第19頁(yè)/共23頁(yè)Therelationshipbetweencomputervisionandothersubjects1）(imageprocessing)Imageprocessingisusuallythetransformationofanimageintoanotherimage;Thecomputervisionsystemtakestheimageastheinputsignalandgeneratesanotherformofoutputsignal.Imageprocessingplaysagreatroleintheearlystageofmachinevisionsystem.2）

(computergraphics)Itusegeometricprimitives(suchaslines,circles)togeneratetheimage,whichbelongstotheimagesynthesis;computervisionisjusttosolvetheoppositeproblem,thatistoestimatethegeometricprimitivesandotherfeaturesfromtheimage,whichbelongtotheimageanalysis.Visualizationandvirtualrealityarecloselyrelatedtothetw