搜索
I.J.Image,GraphicsandSignalProcessing,2018,9,34-41PublishedOnlineSeptember2018inMECS()DOI:10.5815/ijigsp.2018.09.05Copyright©2018MECSI.J.Image,GraphicsandSignalProcessing,2018,9,34-41AVideobasedVehicleDetection,CountingandClassificationSystemSheerazMemonDepartmentofComputerSystemEngineering,MehranUniversityofEngineering&Technology,Jamshoro,Pakistan.Email:sheeraz.memon@faculty.muet.edu.pkSaniaBhatti*,LiaquatA.Thebo**,MirMuhammadB.Talpur**andMohsinA.Memon**DepartmentofSoftwareEngineering,MehranUniversityofEngineering&Technology,Jamshoro,Pakistan.**DepartmentofComputerSystemEngineering,MehranUniversityofEngineering&Technology,Jamshoro,Pakistan.Email:sania.bhatti@faculty.muet.edu.pk,liaquat.thebo@faculty.muet.edu.pk,baqir.talpur@gmail.com,mohsin.memon@faculty.muet.edu.pkReceived:25May2018;Accepted:03July2018;Published:08September2018Abstract—TrafficAnalysishasbeenaproblemthatcityplannershavedealtwithforyears.Smarterwaysarebeingdevelopedtoanalyzetrafficandstreamlinetheprocess.Analysisoftrafficmayaccountforthenumberofvehiclesinanareapersomearbitrarytimeperiodandtheclassofvehicles.Peoplehavedesignedsuchmechanismfordecadesnowbutmostoftheminvolveuseofsensorstodetectthevehiclesi.e.acoupleofproximitysensorstocalculatethedirectionofthemovingvehicleandtokeepthevehiclecount.Eventhoughoverthetimethesesystemshavematuredandarehighlyeffective,theyarenotverybudgetfriendly.Theproblemissuchsystemsrequiremaintenanceandperiodiccalibration.Therefore,thisstudyhaspurposedavisionbasedvehiclecountingandclassificationsystem.ThesysteminvolvescapturingofframesfromthevideotoperformbackgroundsubtractioninorderdetectandcountthevehiclesusingGaussianMixtureModel(GMM)backgroundsubtractionthenitclassifiesthevehiclesbycomparingthecontourareastotheassumedvalues.Thesubstantialcontributionoftheworkisthecomparisonoftwoclassificationmethods.ClassificationhasbeenimplementedusingContourComparison(CC)aswellasBagofFeatures(BoF)andSupportVectorMachine(SVM)method.IndexTerms—Videosurveillance,detection,videoclassification,GaussianMixtureModel,BagofFeatures,SupportVectorMachine.I.INTRODUCTIONTheneedofefficientmanagementandmonitoringofroadtraffichasincreasedinlastfewdecadesbecauseoftheincreaseintheroadnetworks,thenumberandmostimportantlythesizeofvehicles.Intelligenttrafficsurveillancesystemsareveryimportantpartofmoderndaytrafficmanagementbuttheregulartrafficmanagementtechniquessuchaswirelesssensornetworks[1],Inductiveloops[2]andEMmicrowavedetectors[3]areexpensive,bulkyandaredifficulttoinstallwithoutinterruptingthetraffic.Agoodalternativetothesetechniquescanbevideobasedsurveillancesystems[4].Videosurveillancesystems[4-8]havebecomecheaperandbetterbecauseoftheincreaseinthestoragecapabilities,computationalpowerandvideoencryptionalgorithms[9].Thevideosstoredbythesesurveillancesystemsaregenerallyanalyzedbyhumans,whichisatimeconsumingJob.Toovercomethisconstraint,theneedofmorerobust,automaticvideobasedsurveillancesystemshasincreasedinterestinfieldofcomputervision.Theobjectivesofatrafficsurveillancesystemistodetect,trackandclassifythevehiclesbuttheycanbeusedtodocomplextaskssuchasdriveractivityrecognition,lanerecognitionetc.Thetrafficsurveillancesystemscanhaveapplicationsinarangeoffieldssuchas,publicsecurity,detectionofanomalousbehavior,accidentdetection,vehicletheftdetection,parkingareas,andpersonidentification.ATrafficsurveillancesystemusuallycontainstwoparts,hardwareandsoftware.Hardwareisastaticcamerainstalledontheroadsidethatcapturesthevideofeedandthesoftwarepartofthesystemisconcernedwithprocessingandanalyses.Thesesystemscouldbeportablewithamicrocontrollerattachedtothecameraforthereal-timeprocessingandanalysesorjustthecamerasthattransmitthevideofeedtoacentralizedcomputerforfurtherprocessing.II.RELATEDWORKVariousapproachesweremadetodevelopsuchsystemsthatcandetect,countandclassifythevehiclesandcanbeusedfortrafficsurveillanceinintelligenttransportationsystems.Thissectioncoversthediscussionaboutsuchsystemsandtheknowledgeaboutthemethodsusedtodevelopsuchsystems.Tursun,MandAmrulla,G[4]proposedavideobasedreal-timevehiclecountingsystemusingoptimizedvirtualloopmethod.TheyusedrealtimetrafficsurveillancecamerasdeployedoverroadsandcomputehowmanyAVideobasedVehicleDetection,CountingandClassificationSystem35Copyright©2018MECSI.J.Image,GraphicsandSignalProcessing,2018,9,34-41vehiclespasstheroad.Inthissystemcountingiscompletedinthreestepsbytrackingvehiclemovementswithinatrackingzonecalledvirtualloop.AnothervideobasedvehiclecountingsystemwasproposedbyLei,M.,etal.[5].Inthissystemsurveillancecameraswereusedandmountedatrelativelyhighplacetoacquirethetrafficvideostream,theAdaptivebackgroundestimationandtheGaussianshadoweliminationarethetwomainmethodsthatwereusedinthissystem.Theaccuracyrateofthesystemdependsonthevisualangleandabilitytoremoveshadowsandghosteffects.Thesystem’sincompetencytoclassifyvehicletypeisthecorelimitationofthesystem.Basetal.proposedavideoanalysismethodtocountvehicles[10]basedonanadaptiveboundingboxsizetodetectandtrackvehiclesinaccordancewithe