Määritelmä
Computervisionisasimulationofbiologicalvisionusingcomputersandrelatedequipment.Itsmaintaskistoobtainthree-dimensionalinformationofthecorrespondingscenebyprocessingthecollectedpicturesorvideos,justlikehumansandmanyotherkindsofcreaturesdoeveryday.
Computervisionisasubjectofhowtousecamerasandcomputerstoobtainthedataandinformationweneedaboutthesubject.Yläosautitvividly,itistoinstalleyes(cameras)andbrains(algorithms)onthecomputersothatthecomputercanperceivetheenvironment.TheChineseidiom"seeingisbelieving"andthewesternsaying"Onepictureisworthtenthousandwords"expresstheimportanceofvisiontomankind.Itisnotdifficulttoimaginehowbroadtheapplicationprospectsofmachineswithvisioncanbe.
Computervisionisnotonlyanengineeringfield,butalsoachallengingandimportantresearchfieldinthescientificfield.Computervisionisacomprehensivediscipline,ithasattractedresearchersfromvariousdisciplinestoparticipateinitsresearch.Theseincludecomputerscienceandengineering,signalprocessing,physics,appliedmathematicsandstatistics,neurophysiologyandcognitivescience.
Analyysi
Visionisanintegralpartofvariousintelligent/autonomoussystemsinvariousapplicationfields,suchasmanufacturing,inspection,documentanalysis,medicaldiagnosis,andmilitary.Becauseofitsimportance,someadvancedcountries,suchastheUnitedStates,listcomputervisionresearchasamajorbasicprobleminscienceandengineeringthathasawide-rangingimpactoneconomyandscience,theso-calledgrandchallenge.Thechallengeofcomputervisionistodevelopvisioncapabilitiescomparabletohumansforcomputersandrobots.Machinevisionrequiresimagesignals,textureandcolormodeling,geometricprocessingandreasoning,andobjectmodeling.Acapablevisionsystemshouldtightlyintegratealltheseprocesses.Asadiscipline,computervisionbeganintheearly1960s,butmanyimportantadvancesinthebasicresearchofcomputervisionweremadeinthe1980s.Computervisioniscloselyrelatedtohumanvision.Acorrectunderstandingofhumanvisionwillbeverybeneficialtotheresearchofcomputervision.Forthiswewillfirstintroducehumanvision.
Periaate
Computervisionistheuseofvariousimagingsystemsinsteadofvisualorgansasinput-sensitivemeans,andcomputersinsteadofthebraintocompleteprocessingandinterpretation.Theultimateresearchgoalofcomputervisionistoenablecomputerstoobserveandunderstandtheworldthroughvisionlikehumans,andhavetheabilitytoadapttotheenvironmentautonomously.Agoalthatcanonlybeachievedafterlong-termefforts.Therefore,beforeachievingthefinalgoal,themid-termgoalofpeople'seffortsistoestablishavisionsystemthatcancompletecertaintasksbasedonacertaindegreeofintelligencewithvisualsensitivityandfeedback.Forexample,animportantapplicationareaofcomputervisionisthevisualnavigationofautonomousvehicles.Thereisnoconditiontorealizeasystemthatcanrecognizeandunderstandanyenvironmentandcompleteautonomousnavigationlikehumans.Therefore,theresearchgoalofpeople'seffortsistoachieveavisualassisteddrivingsystemthathasroadtrackingcapabilitiesonexpresswaysandcanavoidcollisionswithvehiclesinfront.Thepointtobepointedouthereisthatinthecomputervisionsystem,thecomputerreplacesthehumanbrain,butitdoesnotmeanthatthecomputermustcompletetheprocessingofvisualinformationaccordingtothemethodofhumanvision.Computervisioncanandshouldprocessvisualinformationaccordingtothecharacteristicsofthecomputersystem.However,thehumanvisualsystemisbyfarthemostpowerfulandcompletevisualsystemknowntopeople.Asyouwillseeinthefollowingchapters,thestudyofhumanvisualprocessingmechanismswillprovideinspirationandguidanceforcomputervisionresearch.Therefore,thecomputerinformationprocessingmethodisusedtostudythemechanismofhumanvisionandestablishthecalculationtheoryofhumanvision.ResearchinthisareaiscalledComputationalVision.Computationalvisioncanbeconsideredasaresearchfieldincomputervision.
Liittyvät
Therearemanydisciplineswhoseresearchgoalsaresimilartoorrelatedtocomputervision.Thesesubjectsincludeimageprocessing,patternrecognitionorimagerecognition,sceneanalysis,imageunderstanding,etc.Computervisionincludesimageprocessingandpatternrecognition.Inaddition,italsoincludesthedescriptionofspatialshapes,geometricmodeling,andtheprocessofrecognition.Realizingimageunderstandingistheultimategoalofcomputervision.
Kuvankäsittely
Kuvankäsittelytechnologyconvertstheinputimageintoanotherimagewithdesiredcharacteristics.Forexample,theoutputimagecanbeprocessedtohaveahighersignal-to-noiseratio,orenhancedprocessingcanbeusedtohighlightthedetailsoftheimagetofacilitateinspectionbytheoperator.Incomputervisionresearch,imageprocessingtechnologyisoftenusedforpreprocessingandfeatureextraction.
Hahmontunnistus
Hahmontunnistustechnologydividesimagesintopredeterminedcategoriesbasedonthestatisticalcharacteristicsorstructuralinformationextractedfromtheimage.Forexample,textrecognitionorfingerprintrecognition.Incomputervision,patternrecognitiontechnologyisoftenusedtoidentifyandclassifycertainpartsofanimage,suchassegmentedregions.
Kuvan ymmärtäminen
Givenanimage,theimageunderstandingprogramnotonlydescribestheimageitself,butalsodescribesandinterpretsthescenerepresentedbytheimage,inordertomakeananalysisofthecontentrepresentedbytheimage.Decide.Intheearlydaysofartificialintelligencevisionresearch,thetermsceneanalysiswasoftenusedtoemphasizethedifferencebetweentwo-dimensionalimagesandthree-dimensionalscenes.Inadditiontocompleximageprocessing,imageunderstandingalsorequiresknowledgeaboutthephysicallawsofsceneimagingandknowledgerelatedtothecontentofthescene.
Whenestablishingacomputervisionsystem,itisnecessarytousetherelevanttechnologiesintheabovedisciplines,butthecontentofcomputervisionresearchismoreextensivethanthesedisciplines.Theresearchofcomputervisioniscloselyrelatedtotheresearchofhumanvision.Inordertoachievethegoalofestablishingageneral-purposecomputervisionsystemsimilartothehumanvisionsystem,itisnecessarytoestablishacomputertheoryofhumanvision.
Nykyinen tilanne
Theoutstandingfeatureofthecomputervisionfieldisitsdiversityandimperfection.Pioneersinthisfieldcanbetracedbacktoearliertimes,butitwasnotuntilthelate1970swhentheperformanceofcomputerswasimprovedtohandlelarge-scaledatasuchasimagesthatcomputervisionreceivedformalattentionanddevelopment.However,thesedevelopmentsoftenoriginatefromtheneedsofotherdifferentfields,sowhatismeantby"computervisionproblems"hasneverbeenformallydefined.Naturally,thereisnoformulaforhow"computervisionproblems"shouldbesolved.
Nevertheless,peoplehavebeguntomastersomeofthemethodstosolvespecificcomputervisiontasks.Unfortunately,thesemethodsareusuallyonlyapplicabletoagroupofnarrowtargets(suchas:faces,fingerprints,text,etc.),sotheycannotbeWidelyusedindifferentoccasions.
Theapplicationofthesemethodsisusuallyacomponentofsomelarge-scalesystemsthatsolvecomplexproblems(suchasmedicalimageprocessing,qualitycontrolandmeasurementinindustrialmanufacturing).Inmostpracticalapplicationsofcomputervision,computersarepresettosolvespecifictasks.However,methodsbasedonmachinelearningarebecomingmoreandmorepopular.Oncetheresearchofmachinelearningisfurtherdeveloped,thefuture"generalpurpose"computervisionapplicationsmaybeabletocometrue.
Oneofthemainissuesstudiedbyartificialintelligenceis:howtomakethesystemhave"planning"and"decision-makingcapabilities"?Soastomakeitcompleteaspecifictechnicalaction(forexample:movearobotthroughaspecificenvironment).Thisproblemiscloselyrelatedtothecomputervisionproblem.Here,thecomputervisionsystemactsasaperceptron,providinginformationfordecision-making.Otherresearchdirectionsincludepatternrecognitionandmachinelearning(whichalsobelongtothefieldofartificialintelligence,buthaveanimportantconnectionwithcomputervision).Asaresult,computervisionisoftenregardedasabranchofartificialintelligenceandcomputerscience.
Physicsisanotherfieldthathasanimportantconnectionwithcomputervision.
Thegoalofcomputervisionistofullyunderstandtheelectromagneticwaves-mainlyvisiblelightandinfraredlight-theimageformedbythereflectionofthesurfaceoftheobject,andthisprocessisbasedonopticalphysicsandsolid-statephysics.Somecutting-edgeimageperceptionsystemswillevenbeappliedtoquantummechanicstheorytoanalyzetherealworldrepresentedbyimages.Atthesametime,manymeasurementproblemsinphysicscanalsobesolvedbycomputervision,suchasfluidmotion.Becauseofthis,computervisioncanalsobeseenasanextensionofphysics.
Anotherimportantfieldisneurobiology,especiallythepartofthebiologicalvisualsystem.
Throughoutthe20thcentury,humanshaveconductedextensivestudiesontheeyes,neurons,andbraintissuesofvariousanimalsrelatedtovisualstimulation.Thesestudieshaveledtosome"natural"Thedescriptionofhowthevisualsystemworks(althoughitisstillabitrough)hasalsoformedasub-fieldofcomputervision-peopletrytobuildartificialsystemsthatcansimulatethevisualoperationsoflivingbeingswithvaryingdegreesofcomplexity.Atthesametime,inthefieldofcomputervision,somemethodsbasedonmachinelearningalsorefertosomebiologicalmechanisms.
Anotherrelatedfieldofcomputervisionissignalprocessing.Manyprocessingmethodsrelatedtounitvariablesignals,especiallytheprocessingoftime-varyingsignals,cannaturallybeextendedtotheprocessingmethodsofbinaryvariablesignalsormultivariatesignalsincomputervision.However,duetotheuniquepropertiesofimagedata,manymethodsdevelopedincomputervisioncannotfindacorrespondingversionintheunitsignalprocessingmethod.Oneofthemaincharacteristicsofthesemethodsistheirnon-linearityandthemulti-dimensionalityofimageinformation.Theabovetwopoints,aspartofcomputervision,formaspecialresearchdirectioninsignalprocessing.
Inadditiontothefieldsmentionedabove,manyresearchtopicscanalsobetreatedaspurelymathematicalproblems.Forexample,manyproblemsincomputervisionarebasedonstatistics,optimizationtheory,andgeometry.
Howtoimplementexistingmethodsthroughvarioussoftwareandhardware,orhowtomodifythesemethods,soastoobtainreasonableexecutionspeedwithoutlosingsufficientaccuracy,isthemainissueinthefieldofcomputervisiontoday.Subject.
Sovellus
Mankindisenteringtheinformationage,andcomputerswillincreasinglyenteralmostallfields.Ontheonehand,morepeoplewithoutprofessionalcomputertrainingalsoneedtousecomputers.Ontheotherhand,thefunctionsofcomputersaregettingstrongerandstronger,andthemethodsofusingthemaregettingmoreandmorecomplicated.Thiscreatesasharpcontradictionbetweentheflexibilityofpeopleinconversationandcommunicationandthestrictnessandrigidityrequiredwhenusingcomputers.Humanscanexchangeinformationwiththeoutsideworldthroughvision,hearing,andlanguage,andcanexpressthesamemeaningindifferentways.However,computersarerequiredtowriteprogramsstrictlyinaccordancewithvariousprogramminglanguages,sothatcomputerscanrun.Inordertoenablemorepeopletousecomplexcomputers,itisnecessarytochangethepastsituationwherepeopleadapttocomputersandmemorizecomputerusagerulesbyrote.Instead,letthecomputeradapttopeople'shabitsandrequirements,andexchangeinformationwithpeopleinthewaypeopleareusedto,thatis,letthecomputerhavetheabilitytosee,hear,andspeak.Atthistime,thecomputermusthavetheabilityoflogicalreasoninganddecision-making.Acomputerwiththeabovecapabilitiesisanintelligentcomputer.
Intelligentcomputersnotonlymakecomputersmoreconvenientforpeopletouse,butatthesametime,ifsuchcomputersareusedtocontrolvariousautomationdevices,especiallyintelligentrobots,theseautomationsystemsandintelligentrobotscanadapttotheenvironment,andTheabilitytomakedecisionsindependently.Thiscanreplacepeople'sheavyworkonvariousoccasions,orreplacepeopletocompletetasksinvariousdangerousandharshenvironments.
Sovellussrangefromtasks,suchasindustrialmachinevisionsystems,forexample,inspectionofbottlesontheproductionlinetoacceleratethrough,researchintoartificialintelligenceandcomputersorrobots,whichcanunderstandtheworldaroundthem.Thereisasignificantoverlapinthefieldsofcomputervisionandmachinevision.Computervisioninvolvesthecoretechnologyusedinautomatedimageanalysisinmanyfields.Machinevisionusuallyreferstoaprocessthatcombinesautomaticimageanalysiswithothermethodsandtechnologiestoprovideautomaticdetectionandrobotguidanceinindustrialapplications.Inmanycomputervisionapplications,computersarepre-programmedtosolvespecifictasks,butlearning-basedmethodsarenowbecomingmoreandmorecommon.Examplesofcomputervisionapplicationsincludethoseusedinsystems:
(1) Prosessin hallinta, kuten teollisuusrobotti;
(2) Navigointi, esimerkiksi itsenäisten autojen tai mobiilirobottien avulla;
(3)Havaitut tapahtumat, kuten videovalvonta ja ihmisten laskenta;
(4)Tiedon järjestäminen, esimerkiksi kuvien ja kuvasekvenssien hakemistotietokannat;
(5)Modelingobjectsorenvironments,suchasmedicalimageanalysissystemsorterrainmodels;
(6) Vuorovaikutus esimerkiksi syötettäessä laitteeseen tietokoneen ja ihmisen vuorovaikutusta varten;
(7) Automaattinen tunnistus,esimerkiksi valmistussovellukset.
Themostprominentapplicationareasaremedicalcomputervisionandmedicalimageprocessing.Thefeatureinformationofthisareaisextractedfromtheimagedataforthepurposeofmedicaldiagnosisofthepatient.Usually,theimagedataisintheformofmicroscopeimages,X-rayimages,angiographyimages,ultrasoundimagesandtomographicimages.Anexampleoftheinformationthatcanbeextractedfromsuchimagedataisthedetectionoftumors,atherosclerosisorothermalignantchanges.Itcanalsobethesizeoftheorgan,bloodflow,etc.Thisfieldofapplicationalsosupportsthemeasurementofmedicalresearchbyprovidingnewinformation,forexample,onthestructureofthebrain,oraboutthequalityofmedicaltreatment.Theapplicationofcomputervisioninthemedicalfieldalsoincludesenhancingimagesthatareinterpretedbyhumans,suchasultrasoundimagesorX-rayimages,toreducetheeffectsofnoise.
Thesecondapplicationareaofcomputervisionisinindustry,sometimescalledmachinevision,whereinformationisextractedtosupportthepurposeofthemanufacturingprocess.Anexampleisqualitycontrol,wheretheinformationorfinalproductisautomaticallydetectedinordertofinddefects.Anotherexampleisthatthepositionanddetailorientationbeingpickeduparemeasuredbytheroboticarm.Machinevisionisalsousedextensivelyintheprocessofagriculture,frombulkmaterials,thisprocessiscalledtheremovalofunwantedthings,opticalsortingoffood.
Militaryapplicationsareprobablyoneofthelargestareasofcomputervision.Themostobviousexampleisthedetectionofenemysoldiersorvehiclesandmissileguidance.Moreadvancedsystemsguidethemissiletotheareawherethemissileissent,ratherthanaspecifictarget,andmakeaselectionwhenthemissilereachesthetargetintheareabasedonlocallyacquiredimagedata.Modernmilitaryconcepts,suchas"battlefieldperception",meanthatvarioussensors,includingimagesensors,provideawealthofrelevantcombatscenariosthatcanbeusedtosupportstrategicdecision-makinginformation.Inthiscase,automaticdataprocessingisusedtoreducecomplexityandfuseinformationfrommultiplesensorstoimprovereliability.
Anewerapplicationareaisautonomousvehicles,whichincludediving,landvehicles(smallrobotswithwheels,carsortrucks),aerialworkvehiclesandunmannedaerialvehicles(UAV).Thelevelofautonomyrangesfromcompletelyindependent(unmanned)vehiclestocars,wherecomputervision-basedsystemssupportdriverprogramsorexperimentsindifferentsituations.Afullyautonomouscarusuallyusescomputervisiontonavigatewhenitknowswhereitis,ortheenvironmentusedforproduction(mapSLAM)andfordetectingobstacles.Itcanalsobeusedtodetectspecificeventsforspecifictasks,forexample,aUAVlookingforforestfires.Examplesofsupportsystemsarecarsinobstaclewarningsystems,andautonomouslandingsystemsforaircraft.Severalautomakershavedemonstratedthesystem'sautonomousdrivingofcars,butthetechnologyhasnotreachedacertainlevelbeforeitcanbeputonthemarket.Thereareplentyofexamplesofmilitaryautonomousmodels,fromadvancedmissiles,unmannedaerialvehiclesforreconnaissancemissionsormissileguidance.Spaceexplorationisalreadyusingcomputervision,autonomousvehiclessuchasNASA’sMarsExplorationRoverandtheEuropeanSpaceAgency’sExoMarsMarsRover.
Muita sovellusalueita ovat:
(1)Elokuvat ja lähetykset, jotka tukevat visuaalisten tehosteiden tuotantoa, esimerkiksi kameran seurantaa (liikkeensovitus).
(2) Valvonta.
Samankaltaisuudet ja eroavaisuudet
Computervision,imageprocessing,imageanalysis,robotvisionandmachinevisionarecloselyrelateddisciplines.Ifyouopenthetextbookswiththeabovenames,youwillfindthattheyhaveaconsiderableoverlapintechnologyandapplicationareas.Thisshowsthatthebasictheoriesofthesedisciplinesareroughlythesame,anditevenmakespeoplesuspectthattheyarethesamedisciplineswithdifferentnames.
However,variousresearchinstitutions,academicjournals,conferences,andcompaniesoftenclassifythemselvesasaparticularfield,soavarietyofcharacteristicsthatdistinguishthesedisciplineshavebeenbroughtup.Amethodofdistinctionwillbegivenbelow,althoughitcannotbesaidthatthismethodofdistinctioniscompletelyaccurate.
Theresearchobjectofcomputervisionismainlyathree-dimensionalscenemappedtoasingleormultipleimages,suchasthereconstructionofathree-dimensionalscene.Theresearchofcomputervisionislargelyfocusedonthecontentoftheimage.
Theresearchobjectsofimageprocessingandimageanalysisaremainlytwo-dimensionalimages,whichrealizeimagetransformation,especiallyforpixel-leveloperations,suchasimagecontrastimprovement,edgeextraction,denoisingandgeometrictransformationssuchasimagerotation.Thisfeatureshowsthattheresearchcontentofimageprocessingorimageanalysishasnothingtodowiththespecificcontentoftheimage.
Machinevisionmainlyreferstothevisualresearchintheindustrialfield,suchasthevisionofautonomousrobots,andthevisionforinspectionandmeasurement.Thisshowsthatinthisfield,throughsoftwareandhardware,imageperceptionandcontroltheoryisoftencloselycombinedwithimageprocessingtoachieveefficientrobotcontrolorvariousreal-timeoperations.
Hahmontunnistususesvariousmethodstoextractinformationfromsignals,mainlyusingstatisticaltheories.Oneofthemaindirectionsinthisfieldistoextractinformationfromimagedata.
Thereisanotherfieldcalledimagingtechnology.Theinitialresearchcontentinthisfieldismainlytomakeimages,butsometimesalsoinvolvesimageanalysisandprocessing.Forexample,medicalimagingincludesalargenumberofimageanalysisinthemedicalfield.
Forallthesefields,apossibleprocessisthatyouworkinacomputervisionlaboratory,youareengagedinimageprocessingatwork,andfinallysolvetheproblemsinthefieldofmachinevision,andthenpublishyourresultsinAtthemeetingofpatternrecognition.
Ongelmia
Almosteveryspecificapplicationofcomputervisiontechnologymustsolveaseriesofthesameproblems.Theseclassicproblemsinclude:
Tunnustus
Acomputervision,imageprocessingandmachinevisioncommonclassicproblemistodeterminewhetherasetofimagedatacontainsaspecificObject,imagefeatureormovementstate.Thisproblemcanusuallybesolvedautomaticallybyamachine,butsofar,thereisnosinglemethodthatcandetermineawiderangeofsituations:recognizeanyobjectinanyenvironment.Theexistingtechnologycanandcanonlywellsolvetherecognitionofspecifictargets,suchassimplegeometricpatternrecognition,facerecognition,printedorhandwrittendocumentrecognition,orvehiclerecognition.Andtheserecognitionsneedtohavespecifiedlighting,backgroundandtargetposturerequirementsinaspecificenvironment.
Generalrecognitionhasevolvedintoseveralslightlydifferentconceptsondifferentoccasions:
Tunnustus(narrowsense):Foroneormorepre-definedorlearnedObjectsorobjectsarerecognized,andtheirtwo-dimensionalpositionorthree-dimensionalpostureisusuallyprovidedduringtherecognitionprocess.
Identification:Identifythesingleobjectitself.Forexample:therecognitionofacertainface,therecognitionofacertainfingerprint.
Monitoring:Discoverspecificsituationcontentfromimages.Forexample:thediscoveryofabnormalskillsincellsortissuesinmedicine,andthediscoveryofpassingvehiclesbytrafficmonitoringequipment.Monitoringisoftentodiscoverspecialareasintheimagethroughsimpleimageprocessing,whichprovidesastartingpointforsubsequentmorecomplexoperations.
Useita tiettyjä sovellussuuntia tunnistettu:
Content-basedimageextraction:Findallpicturescontainingspecifiedcontentinahugeimagecollection.Thespecifiedcontentcantakemanyforms,suchasaredroughlycircularpattern,orabicycle.Thesearchforthelatterkindofcontenthereisobviouslymorecomplicatedthantheformer,becausetheformerdescribesalow-levelintuitivevisualfeature,whilethelatterinvolvesanabstractconcept(orhigh-levelvisualfeature).Thatis,"bicycle",theobviouspointisthattheappearanceofthebicycleisnotfixed.
Poseevaluation:Evaluationofthepositionordirectionofanobjectrelativetothecamera.Forexample:theassessmentofthepostureandpositionoftheroboticarm.
Opticalcharacterrecognitionrecognizesanddiscriminatesprintedorhandwrittentextinanimage,andtheusualoutputistoconvertitintoaneasy-to-editdocumentform.
Liike
Themonitoringofobjectmotionbasedonsequenceimagesincludesmanytypes,suchas:
Selfmotion:monitorthethree-dimensionalrigidmotionofthecamera.
Kuvanseuranta: Seuraa liikkuvia esineitä.
SceneReconstruction
Giventwoormoreimagesoravideoofascene,scenereconstructionseekstobuildacomputermodel/three-dimensionalmodelofthescene.Thesimplestcaseistogenerateasetofpointsinthree-dimensionalspace.Inmorecomplexsituations,acompletethree-dimensionalsurfacemodelwillbebuilt.
Kuvan palauttaminen
Thegoalofimagerestorationistoremovenoiseintheimage,suchasinstrumentnoise,blur,etc.
Järjestelmä
Thestructureofthecomputervisionsystemlargelydependsonitsspecificapplicationdirection.Someworkindependentlyandareusedtosolvespecificmeasurementorinspectionproblems;someappearasapartofalargecomplexsystem,suchasworkingwithmechanicalcontrolsystems,databasesystems,andman-machineinterfacedevices.Thespecificimplementationmethodofthecomputervisionsystemisalsodeterminedbyitsfunction-whetheritisfixedinadvanceorisautomaticallylearnedandadjustedduringoperation.However,therearesomefunctionsthatalmosteverycomputersystemneeds:
Kuvan hankinta
Adigitalimageisproducedbyoneormoreimagesensors,hereThesensorcanbeavarietyofphotosensitivecameras,includingremotesensingequipment,X-raytomography,radar,ultrasonicreceivers,andsoon.Dependingonthedifferentperceptrons,thegeneratedpicturecanbeanordinarytwo-dimensionalimage,athree-dimensionalimagegrouporanimagesequence.Thepixelvalueofthepictureoftencorrespondstotheintensityoflightinoneormorespectralbands(grayscaleorcolorimage),butitcanalsoberelatedtovariousphysicaldata,suchasthedepthandabsorbanceofsoundwaves,electromagneticwavesornuclearmagneticresonanceOrreflectivity.
Esikäsittely
Beforeimplementingspecificcomputervisionmethodsontheimagetoextractcertainspecificinformation,oneorsomepreprocessingisoftenusedtomaketheimagemeettherequirementsofsubsequentmethodsRequire.Forexample:
Osanäytteenotto oikeiden kuvankoordinaattien varmistamiseksi;
Smoothdenoisingtofilteroutthedevicenoiseintroducedbythesensor;
ImprovethecontrasttoensuretherealizationRelevantinformationcanbedetected;
Adjustthescalespacetomaketheimagestructuresuitableforlocalapplications.
Ominaisuuksien erottaminen
Extractfeaturesofvariouscomplexityfromtheimage.Forexample:
Line, reunan poisto;
Localizedfeaturepointdetectionsuchascornerdetection,spotdetection;
MorecomplexfeaturesmayberelatedtotheimageThetextureshapeormovementisrelated.
Havaitsemissegmentointi
Intheprocessofimageprocessing,itissometimesnecessarytosegmenttheimagetoextractvaluablepartsforsubsequentprocessing,suchas
seulontaOminaisuuspisteet;
Segmentthepartofoneormorepicturesthatcontainsaspecifictarget.
Edistynyt käsittely
Atthispoint,thedataoftenhasasmallamount,suchasthepartoftheimagethatisconsideredtocontainthetargetobjectafterpreviousprocessing.Theprocessingatthistimeincludes:
Verifywhetherthedataobtainedmeetstheprerequisiterequirements;
Estimatespecificcoefficients,suchasthetarget’sattitudeandvolume;
järjestellä.
Edistynyt käsittelyhasthemeaningofunderstandingimagecontent.Itisahigh-levelprocessingincomputervision.Itismainlybasedonimagesegmentationtounderstandthesegmentedimageblocks,suchasrecognitionandotheroperations..
Vaatimukset
Theinfluenceoflightsourcelayoutneedstobecarefullyconsidered.
Valitse oikeat ryhmät ottaen huomioon suurennus, tila, koko, vääristymä...
Valitse oikea kamera (CCD) ottaen huomioon toiminnot, tekniset tiedot, vakauden, kestävyyden...
Visualsoftwaredevelopmentneedstorelyontheaccumulationofexperience,trymoreandthinkaboutthewaytosolvetheproblem.
Theultimategoalistocontinuouslyimprovetheaccuracyofcreationandshortentheprocessingtime.
loppu.
Konferenssi
Yläosa
ICCV:InternationalKonferenssionComputerVision,InternationalComputerVisionKonferenssi
CVPR:InternationalKonferenssionComputerVisionandPatternTunnustus,InternationalKonferenssionComputerVisionandPatternTunnustus
ECCV:EuropeanKonferenssionComputerVision,EuropeanKonferenssionComputerVision
Paremmin
ICIP:InternationalKonferenssionImageProcessing,InternationalKonferenssionImageProcessing
BMVC:BritishMachineVisionKonferenssi,BritishMachineVisionKonferenssi
ICPR:InternationalKonferenssionPatternTunnustus,InternationalKonferenssionPatternTunnustus
ACCV:AsianKonferenssionComputerVision,AsianKonferenssionComputerVision
Journal
Yläosa
PAMI:IEEETransactionsonPatternAnalyysiandMachineIntelligence,IEEEPatternAnalyysiJournalofMachineIntelligence
IJCV:InternationalJournalonComputerVision,InternationalJournalofComputerVision
Paremmin
TIP:IEEETransactionsonImageProcessing,IEEEImageProcessingMagazine
CVIU:ComputerVisionandImageUnderstanding,ComputerVisionandImageUnderstanding
PR: Pattern Tunnustus, Pattern Tunnustus
PRL:PatternTunnustusLetters,PatternTunnustusExpress