资源描述:
《weather recognition based on images captured》由会员上传分享,免费在线阅读,更多相关内容在教育资源-天天文库。
1、WeatherRecognitionBasedonImagesCapturedbyVisionSysteminVehicleXunshiYan1,YupinLuo1,andXiaomingZheng21TsinghuaNationalLaboratoryforInformationScienceandTechnology(TNList),DepartmentofAutomation,TsinghuaUniversity,Beijing100084,China2INFTechnologies,Ltd.,Beijing100086,C
2、hinayanxs06@mails.thu.edu.cn,luo@tsinghua.edu.cn,zheng@tsinghua.edu.cnAbstract.Weatherrecognitioniswidelyrequiredinmanyareas,anditisalsoachallengingandbrand-newsubject.Thispaperproposesanapproachtorecognizeweatherbasedonimagescapturedbyin-vehiclevisionsystem.Webringth
3、reegroupsoffeatures,includinghistogramofgradientamplitude,HSVcolorhistogram,roadinformation,andemployanalgorithmbasedonRealAdaBoost,makinguseofthecategorystruc-turetoachievethetaskofclassification.Experimentsconfirmsuperiorperformancesonourdatasetcollectedfromimagescapt
4、uredbyvisionsystem.Keywords:Weatherrecognition,Visionsystem,RealAdaBoost,HSVcolorspace,Categorystructure.1IntroductionComputervisionsystemhasachievedgreatsuccessinmanyareas,suchassurveillance,navigation,driverassistancesystem.However,thecamerasexposedoutsideareeasilyi
5、nfluencedbybadweather.Forexample,pedestriandetectionsysteminvehiclecouldnotworkatallwhenraindropfallsonthecamera,evenresultsseriousfalse-detection.Manyvisionsystemsalsoneedtoresetparameterssuchaslighting,rainwiper,underdifferentweatherconditions.Hence,researchofweatherr
6、ecognitioninvisionsystemisinurgentdemand.Weatherrecognitionisabrand-newsubjectandonlyafewofpreviousworkhasaddressedthisissue.GargandNaya[1]inColumbiaUniversityfocusondetectingandremovingrainstreaksfromvideos.Theideacomesfrommovingobjectdetection.Itmakesadifferencebetwe
7、enthetwoadjacentframes,andcangivesomeperfectresultsundercertainscenarios,butitishardtosatisfythedynamicbackgroundorthesituationofraindropsadherentonthecam-era.KurihataandandTakahashi[2][3]inNagoyaUniversitycollectlargemountofraindroppatches,utilizeprincipalcomponentan
8、alysistomakearaindroptemplate.Theysearchtheglobalorpartofimagesbycomputingthesimilar-itybetweentheraindroptemplateandthepatc