Konenäkö

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,animportantapplicationareaof​​computervisionisthevisualnavigationofautonomousvehicles.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.

Thesecondapplicationareaof​​computervisionisinindustry,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.

Computer vision

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

Related Articles
TOP