Stereo Vision Introduction and Applications | Teledyne FLIR

Each pixel in a digital camera image collects light that reaches the camera along a 3D ray. If a feature in the world can be identified as a pixel location ... Icon-ContactSalesShapeGridViewExportedfromStreamlineApp(https://app.streamlineicons.com)ModelSelectorModelSelector DownloadPDF-Stereo-Applications Overview Stereovisionisanimagingtechniquethatcanprovidefullfieldofview3Dmeasurementsinanunstructuredanddynamicenvironment.Thefoundationofstereovisionissimilarto3Dperceptioninhumanvisionandisbasedontriangulationofraysfrommultipleviewpoints.Eachpixelinadigitalcameraimagecollectslightthatreachesthecameraalonga3Dray.Ifafeatureintheworldcanbeidentifiedasapixellocationinanimage,weknowthatthisfeatureliesonthe3Drayassociatedwiththatpixel. Ifweusemultiplecameras,wecanobtainmultiplerays. Theintersectionoftheseraysisthe3Dlocationofthefeature. Inpractice,thekeyproblemstosolveinstereovisionare: Identifywhichpixelsinmultipleimagesmatchthesameworldfeature. Thisisknownasthecorrespondenceproblem. Identifyforeachpixelintheimagethecorrespondingrayin3Dspace. Thisisknownasthecalibrationproblem,andrequiresaccuratecalibrationofthecameraopticalparametersandphysicallocation. Thecorrespondenceproblemissolvedthroughimageprocessingsoftware. Dependingontheapplication,thealgorithmusedmaysolvecorrespondencesforonlyasparsesetoffeaturesintheimage(feature-basedalgorithms),orattempttofindcorrespondencesforeverypixelintheimage(densestereoalgorithms).  Cameracalibrationisdonebeforeusingthestereorig. Fordensestereoalgorithms,typicallytheimagesfromthecamerasmustberemappedtoanimagethatfitsapin-holecameramodel. Thisremappedimageiscalledtherectifiedimage. Forlenseswithbarreldistortion,straightlinesintheworldwillappearcurvedintheimage. Intherectifiedimage,however,barreldistortionisremovedandstraightlineswillappearstraight.  Figure1:Wide-anglelenswithbarreldistortion Figure2:Rectifiedpin-holemodelimage Theaccuracyofthe3Dresultsofstereomatchingdependsuponmanyfactorssuchasimagetexture,imageresolution,lensfocallengthandtheseparationbetweencameras.Increaseincameraseparationornarrowerfield-of-viewlensesimprovetheaccuracyatlongrange.Higherimageresolutionincreasestheaccuracyoftheresultsbutalsomayincreasetheprocessingtime.   StereoCamerasBy FLIR Wehave madestereovisionpracticalforavarietyofapplicationareasbyprovidinghardwareandsoftwarepackagesthatincludecompletestereoprocessingsupport–fromimagecorrectionandalignmenttodensecorrelation-basedstereomapping. Bumblebee®stereovisioncamerasarefactorycalibratedandcomewithacompletestereovisionsoftwarepackagecalledtheTriclops™SDK.TheBumblebee®2usestwoCCDimagesensorstoprovideabalancebetween3Ddataquality,processingspeed,sizeandprice.Itcanoutput640x480imagesat48FPS,or1024x768imagesat20FPS,viaitsIEEE-1394(FireWire)interface.TheBumblebee®XB3isathree-CCDmulti-baselinestereocameradesignedforimprovedflexibilityandaccuracy.Ituses1.3MPsensorswithtwobaselines—12cmand24cm—andcanoutput1280x960imagesat16FPSviaits800Mb/sIEEE-1394binterface. TheTriclopsStereoSDKincludedwiththeBumblebee2andBumbleebeeXB3suppliesanassembly-optimizedfast-correlationstereocorethatperformsfastSumofAbsoluteDifferences(SAD)stereocorrelation.Thismethodisknownforitsspeed,simplicityandrobustness,andgeneratesdensedisparityimages.TheTriclopsSDKprovidesflexibleaccesstoallimagestagesinthestereoprocessingpipeline,makingitidealforcustomstereoprocessingapproaches.   Applications Stereovisiontechnologyisusedinavarietyofapplications,includingpeopletracking,mobileroboticsnavigation,andmining.Itisalsousedinindustrialautomationand3Dmachinevisionapplicationstoperformtaskssuchasbinpicking,volumemeasurement,automotivepartmeasurementand3Dobjectlocationandidentification. Stereovisionisalsobecomingthetechnologyofchoiceforrangesensinginmobileroboticsnavigation.TheBumblebee2wasselectedastheprimaryrangesensingmethodforthePrincetonvehicle,“Prowler”,intheDARPAUrbanChallenge,anautonomousvehicleresearchanddevelopmentprogramthathasroboticvehiclescompetinginaraceinacomplexurbanenvironment. Whileotherteamsreliedonamultitudeofoftenextremelyexpensivesensorstoguidetheirroboticcarsthroughthecourse,Prowlerusedonlythreestereovideocamerastoseeandavoidobstacles. “WechosetheBumblebeecamerasfromFLIR becausetheyprovidealotofqualitydataforlowcost,”explainsGordonFranken,teamspokesmanforthePrincetonteam.“Thecamerascanseeoutto60metersandwewereabletowritejustoneobstacledetectionalgorithmandapplyittoallthecameras.”  Frankenadds“Wealsolikethecamerasbecausetheyarepassivesensorsanddon’trelyonbeamingoutlaserorradarwaves. Inaddition,thecameraswouldn’tbeaffectedbystraylaserorradarbeamscomingfromacompetingvehicle.” Randombinpicking(RBP)isanotherapplicationthatbenefitsfromtheuseof3Dvision.RBPrequiresarobotarmtolocateandpickoutspecificitemsfromamixtureofpartshaphazardlypiledinacontainer. Automationofthistaskisfraughtwithdifficultiesthatareprimarilysensingchallenges. Partsneedtobeidentifiedfromanyviewingangle,withoverlap,occlusions,lightingvariabilityandshadowsmakingtheproblemmoredifficult. Wherestereovisionprovidesthekeyadvantage,however,isinfurtherdistinguishingthe3Denvironmentofapartanddeterminingwhetherapartisfreetobegrasped. Thisportionofthetaskiscommonlysolvedwithexpensivelaserscanners,towhichstereoisacosteffectivealternative. Artículosrelacionados Aprendizajeprofundo TroubleshootingNeuralNetworkConversionErrors Lealahistoria Basedeconocimientos ConfiguringGrasshopperGRAS-50S5tooutput5MPimagesat15FPS. Lealahistoria Basedeconocimientos TroubleshootingproblemswhenrunningmultipleUSB2.0camerasonthesamebus Lealahistoria × Confirmarubicación Singapore Mexico