基于動(dòng)態(tài)特征的視頻監(jiān)控系統(tǒng)中的前景分類.pdf

1、中國(guó)海洋大學(xué)碩士學(xué)位論文基于動(dòng)態(tài)特征的視頻監(jiān)控系統(tǒng)中的前景分類姓名：沙曉英申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：計(jì)算機(jī)軟件與理論指導(dǎo)教師：董軍宇20090601ForegroundCIassificationBasingonActiveFeaturesforVisuaISurveiIIanceAbstractThesubjectisaimedatthetechnologyofmovingtargetrecognitionintrafficscenes

2、，whichisoneofthekeytechnologyofintelligentvideosurveillancesystemAtpresent，ITSismostlyfocusedonthedetectionandrecognitionofvehicle，anditiseitherconsideredwithoutpedestrianandbicycle／motorcycleorfocusedalittleonthemTherec

3、ognitionandclassificationofvehicle，bicycle／motorcycleandpedestrianisacomplexprocessThenormaltemplatematchingalgorithmforrecognitionneedsa10tofcomputationsSOthatitCallhardlysatisfytherealtimeandeffectiverequirementsinsurv

4、eillanceTosolvethisquestion，wepresentsanintegratedframeworkintegratingactivetemplatematchingalgorithmandcontextinformationtecognizethemovingtargetThisframeworkcallbeanindependentmoduleembeddedintoavisualsurveillancesyste

5、mTheactivetemplatesareintheformofadictionaryofactivehaarfeaturesbases)，whichareallowedtoslightlyshiftatdifferentlocationsandorientationsTheycanbelearnedforeachobjecttypefromasmallsetofpositivesamplesthatroughlyalignedWit

6、htheselearneddeformabletemplates，themovingforegroundssubtractedfrombackgroundmodelarerecognizedthroughsearchingmaximummatchinglikelihoodToavoidtheexhaustivesearchfortemplatematchingandreducethenoisedisturbance，aschemetoe

7、stimateta唱etsizeandposeatspecificlocationisdevelopedbasedonthecontextualinformationofscenegeometryandtrackingcuesThisframeworkCanbeanindependentmoduleembeddedintoavisualsurveillancesystemItsperformanceandbenefitofusingco