基于视频图像的车型识别系统设计与实现-通信与信息系统专业论文.docx

研究生优秀毕业论文摘要向量机和LibSVM源代码基础上，实现了本系统所用的SVM分类器，并利用己有的训练样本对SVM分类器进行训练，在训练的过程中通过不断测试和修改分类器参数达到更好车辆分类效果。最后，基于论文硬件平台对获取道路车辆视频进行车型识别的系统性能测试，验证论文方法并对最后的测试结果进行分析。关键字：智能交通系统；车辆检测；车型识别；车辆跟踪；支持向量机II万方数据AbstractAbstractVehiclerecognitionisanimportantpartoftheIntelligentTransportationSystem，andvehiclerecognitionmayplayanimportantroleinthetrafficcontrol，automaticparkingcollectionsystemandhighwaytollstations．VehicleRecognitionbasedonvideosequenceshassomecharacteristics，suchasintuitive，economyandrichinformation，thereforetheresearchofvehiclerecognitionbasedonvideosequencesbecomesmoremeaningfulandhasrapidlydeveloped．Meanwhile，谢tllthecontinuousimprovementofthesocialdemand，theaccurateandtimelyvehiclerecognitionsystemwillhaveagreatsocialdemandInthisthesis，wedesignedandimplementedavehicleidentificationsystembasedonvideosequences，whichwasonthebaseofstudyandcomparisonofdomesticandforeignintelligenttransportationapplications．Inthisthesis，wediscussedthefollowingproblems：Firstly,wedesignedavideo—basedimagerecognitionsystemmodels，builtthehardwareplatformoftherecognitionsystemwimcamerasandPC．ThehardwareplatformCanhelptoobtainthereal-timevehiclevideoontheroad，whichWasusedtogetthevideoimageinformationforvehiclerecognition．Astothesoftwareofthissystem，wedesignedandimplementedanidentificationmethodbasedonmulti-featurefusion，accordingtothecomparativeandanalysisofexistingrecognitionalgorithm．Secondpartisvideoimagepre-processingandvehicledetection．Thevideoimagepre—processingWasimplementedbygrayingandfilteringthevideoimagecapturedbycamerasfromtheroad，aimingatremovingnoiseandreducingtheirrelevantinformationofvideoimage．Inthisthesis，weachievedatargetdetectionmethodbasedonbackgroundsubtractionGaussianmixturemodelwhichhadachievedaneffectivelydetectintheapplicationscenariosinthisarticle．Thirdpartisvehiclefeatureextractionandvehicletracking．Toextractthevehiclefeaturetoachievethevehicleclassificationwiththevehicletestresultsabove．Inthisthesis，weappliedamulti-featurefusionmethodincludingvehiclegeometriccharacteri