Efficient and Robust Classification Method

《Efficient and Robust Classification Method》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、528IEEETRANSACTIONSONCIRCUITSANDSYSTEMSFORVIDEOTECHNOLOGY,VOL.15,NO.4,APRIL2005EfficientandRobustClassificationMethodUsingCombinedFeatureVectorforLaneDetectionPangyuJeongandSergiuNedevschi,Member,IEEEAbstract—Theaimofthispaperistodevelopamethodforlow-1)Ana

2、daptive/predefinedgraylevelimagesplittingintocostandaccurateclassificationofhighwaysandruralwaysimagesubimagespixelsforlanedetection.Themethodusesthreemaincomponents:2)TwoclassesK-Meanclassificationofthesubimagesadaptive/predefinedimagesplitting,subimageleve

3、lclassificationusingcombinedandpartiallytunedgrayandtextureandclassmergingbasedonhomogeneitycheckingconditions.Inthefirststep,apreclassificationinroadandnonroadpixelsiscar-basedfeaturevector.riedout,ontheresizedinputimage,usingthedecisiontreemethod.3)Classm

4、ergingbasedonhomogeneitycheckingusingAsaresultofthisfirststepweobtaintheroadreferencefeaturethereferenceroadfeaturevalue.value,andthelane-markingspositionsincaseofhighways.Fortheruralwaysimagesplittingweuseapredefineddivisionmethod,Twodifferentlanedetectio

5、nproceduresareusedinaccor-andforthehighwaysweuseanadaptivedivisionmethodbasedondancetoinputimages:highwaysandruralways.Theanalysisofthedetectedlane-markings.Theproposedclassificationiscarriedthedecisionwindowisusedforautomaticselectionofthelaneoutonthesub

6、imagesusingtheK-meanclassifieronacomposeddetectionprocedurethatisperformedinaccordancetoinputgrayandtexturebasedfeaturevector.Thegrayfeaturevectorisimage.ItwillbeexplainedmoreinSectionIII.fixedinthepreclassificationphase,andthetexturefeaturevectorisonlyupda

7、tedduringtheclassificationisperformed.ThiswayFortheimagesplittingtwokindsofschemesareused:1)theconvergenceismuchfasterandtheclassificationaccuracyispredefinedimagesplittingschemeforruralwaysand2)adaptivebetter.Theresultingroadandnonroadclassesofsubimagesare

8、imagesplittingschemeforhighways.mergedintoaroadandanonroadclassusingahomogeneitycri-terionbasedontheroadreferencefeaturevalue.Next,aforwardInthecaseoftheruralways,weselectthebestsplittingandbackwardmethodisusedtodetectbord