复杂环境下的夜间车辆检测与交通视频昼夜亮度变化模型研究

上海交通大学硕士学位论文复杂环境下的夜间车辆检测与交通视频昼夜亮度变化模型研究姓名：吴海涛申请学位级别：硕士专业：模式识别与智能系统指导教师：方涛20070101HLEPTHLPTHLTPHLEPT96%88%HELPTLogisticLogisticSTUDYOFNIGHTTIMEVIDEOVEHICLEDETECTIONUNDERCOMPLEXENVIRONMENTANDLUMINANCETRANSFORMATIONMODELBETWEENDAYANDNIGHTFORTRAFFICVIDEOABSTRACTIntelligentTransportationSystemisaneffectivemoderntrafficmanagementmethod,whichisbasedontrafficinformationcollectionsystem.Comparedtotraditionaltrafficinformationcollectionsystems,videodetectionsystemsaremoreflexibleandhavemulti-additive-functions.Therefore,videodetectionsystemsarepreferred.However,sincevideodetectionsystemsareverysensibletoroadwayenvironmentandthequalityofvideo,theirdetectionaccuraciesarenothighenoughandtheirdetectionalgorithmsneedtobeimproved.Practicalapplicationshaveindicatedthat,thesesystemshaveevenloweraccuraciesatnight,especiallyatrainynightoronHOVline.Mostpreviousresearchesonnighttimevideovehicledetectionstudiedvehicles’headlights,andtheysimplyassumedthatheadlightshaveregularshapesinvideoimages,buthadnottakenthefollowingcomplexsituationsintoconsiderationadequately:conjunctionsamongheadlights,foglights,indictorsanddecorativelights,presencesofboundarylights,invertedreflectionsofthoselightsinroadseeper.Therefore,thispaperfocusesonnighttimesituationinsteadoftreatingall-weathervideoasawhole,andstudiesitseparately,aimingatcomplicatedsituationssuchasrainyweather,presenceoflargevehiclesand/ortrafficjam.Afteranalyzingdiverseattributesofvideoondifferentroadwayconditionsandvariousrepresentationsofthelightsinvideos,aheadlightextraction,pairingandtrackingalgorithm(HLEPT)ispresented.Thealgorithmextractsheadlightsbybinarizationandmorphologicalmethods.Basedoncameracalibration,thealgorithmfindstheprojectioncoordinatesofheadlightsontheroadwayplanesubjecttocameralens.Accordingtophysicalconstructionandmovementregulationofthevehicle,aseriesofregulationsofheadlightpairing,groupingbyindividualvehicleandtrackingisdesigned.Thealgorithmregardspre-pairingheadlightsandpost-trackingtrajectories(HLTP)methodasprimarymethod,pre-trackingheadlightsandpost-pairingtrajectories(HLPT)methodasaccessorialmethod,andintegratesthetwo.Thatmeans,topairingheadlightsandtrackingtrajectoriesofpairsfirst,andthentoexcludelamp-housesadjacenttothepairs,andtotrackingtheleftlightsandpairingthetrajectoriesatlast.Thealgorithmcalculatesvarioustrafficparametersaccordingtothetrajectories.Experimentresultsindicatethatthisalgorithmisrobustwithlowcomputationalcomplexityandreal-timeperformance,anditsdetectionratioreachesabove96%infinecondition,88%inatrafficjamatrainynightwheninvertedreflectionsoflightsandlargevehiclesexist.SinceHELPTalgorithm’sassumptionofheadlights’featuresandpairingandtrackingregulationsarethesamewithoverheadroadandsuburbroad,itcanalsobeappliedtocollecttrafficinformationontheseroads.Becauseofthewidelydifferenceofilluminationconditionsbetweendayandnight,videovehicledetectionsystemsalwaysemploytwodifferentalgorithms.Thenthequestionsarewhenandhowtoswitchthetwoalgorithms.Inthispaper,amethodispresentedtoanswerthequestions,whichisindependenttocamerahardwareandanalyzeluminancetransformationbyimagesequences.Thetrendofluminancetransformationbetweendayandnightisinvestigated.Afour-parameterLogisticmodelisintroducedtoexpresstheruleofluminancetransformation.Basedontheactuallimitationofluminancechange,thebox-constrainedconditionofparametersisintroduced,andonlineLogisticcurvefittingisexecuted.Bysolvingbox-constrainednon-linearoptimizationproblem,thevaluesofparametersareupdatedreal-timely.Accordingtotheupdatedparameters,thepresentstateofluminancetransformationisidentified,andajudgmentofwhetherornottoswitchdetectionalgorithmsismade.Experimentresultsindicatethatthismodelrecognizeluminancetransformationwell,andcanbeappliedtodecidetoswitchalgorithmsbetweendayandnightinvideovehicledetectionsystems.KEYWORDS:vehicledetection,videodetection,vehicletracking,algorithmswitchingbetweendayandnight,luminancetransformationmodel,non-linearcurvefitting20071232007123200712311.1ITSITSITSITS[1]ITSITS1-1[2]2[3,4]AUTOSCOPE[5]TRAFFICAN,ITERIS[25],PEEK[6][7]1-11.21.2.1[5,8]3AUTOSCOPE,TRAFFICAN,ITERISPEEK[9][10][11][12,13][14][20-22][13][15-17][18][19]1.2.2[23,24]4IterisPeekR.Taktak[29]RitaCucchiara[30,31]R.TaktakWTHTRitaCucchiara1.2.3PeekIterisSetrix[26]SeiichiNAGUMO[27]RachedTaktak[28]SetrixCMOSSeiichiNAGUMO[32]1.35UtahDepartmentofTransportation[4]80%JutaekOhJohnD.LeonardII[6]PeekVideoTrak90024100%R.Taktak[29]RitaCucchiara[30,31]1-11-2HLEPTHeadlightExtraction,PairingandTracking1-1Figure1-1Trafficjamatrainynight6(a)(b)(c)(d)(a)(b)(c)(d)1-2Figure1-2TypicalheadlightsHLEPT1.41-3/71HLPT(HeadlightPairTracking)2HLTP(HeadlightTrajectoryPairing)HLPTHLTP1-3Figure1-3nighttimevideovehicledetectionsystemframework1.512892.1[33,34]1102303452.22.32.310[35]34M×6n≥34M×111111000,...............00TTTTTTTTTnnnTTTnnnPuPPvPPmPPuPPvP−−==−−(2-1)123(,,)Tmmmm=(,,,1)=WixWiyWizi,iiuviim34M×i0z=33M×(2-1)(,,1)WWTiiiPxy=33M×9533M×1323mPumPmPvmP×=××=×(2-2)invM33M×ininvMi1323npxnpnpynp×=××=×(2-3)(,,1)pu