简介：Thispaperpresentsahumandetectionsysteminavision-basedhospitalsurveillanceenvironment.Thesystemiscomposedofthreesubsystems,i.e.backgroundsegmentationsubsystem(BSS),humanfeatureextractionsubsystem(HFES),andhumanrecognitionsubsystem(HRS).ThecodebookbackgroundmodelisappliedintheBSS,thehistogramoforientedgradients(HOG)featuresareusedintheHFES,andthesupportvectormachine(SVM)classificationisemployedintheHRS.Bymeansoftheintegrationofthesesubsystems,thehumandetectioninavision-basedhospitalsurveillanceenvironmentisperformed.Experimentalresultsshowthattheproposedsystemcaneffectivelydetectmostofthepeopleinhospitalsurveillancevideosequences.

简介：Traditionalbackgroundmodelmethodsoftenrequirecomplicatedcomputations,andaresensitivetoilluminationandshadow.Inthispaper,weproposeablock-basedbackgroundmodelingmethod,anduseourproposedmethodtocombinecolorandtexturecharacteristics.Suppressionandrelaxationarethetwokeystrategiestoresistilluminationchangesandshadowdisturbance.Theproposedmethodisquiteefficientandiscapableofresistingilluminationchanges.Experimentalresultsshowthatourmethodissuitableforreal-wordscenesandreal-timeapplications.

简介：Apassiveopticalnetwork(PON)monitoringsystemcombinedlightpulseandfrequencysweeptechniquesisproposedandverifiedinafieldtest.Thelightpulsesurveysovertheallwholenetworkandthefrequencysweepareusedtoinvestigateanyfaultinthelink.Thefieldtestisperformedwith4PONs.EachPONismonitoredat4ports,oneisthesplitterportandtheotherthreearearbitrarychosenmultipleopticalunits(ONUs).AllthetestedPONsaremonitoredinturnsonceperhour.Faultsatthefeederandbranchfiberhavebeenobservedinthisfieldtestandhavebeenanalyzedwiththemonitoringsystem.

简介：Movingobjectextractionandclassificationareimportantproblemsinautomatedvideosurveillancesystems.Abackgroundmodelbasedonregionsegmentationisproposed.AnadaptivesingleGaussianbackgroundmodelisusedinthestableregionwithgradualchanges,andanonparametricmodelisusedinthevariableregionwithjumpingchanges.Ageneralizedagglomerativeschemeisusedtomergethepixelsinthevariableregionandfillinthesmallinterspaces.Atwo-thresholdsequentialalgorithmicschemeisusedtogroupthebackgroundsamplesofthevariableregionintodistinctGaussiandistributionstoacceleratethekerneldensitycomputationspeedofthenonparametricmodel.Inthefeature-basedobjectclassificationphase,thesurveillancesceneisfirstpartitionedaccordingtotheroadboundariesofdifferenttrafficdirectionsandthenre-segmentedaccordingtotheirscenelocalities.Themethodimprovesthediscriminabilityofthefeaturesineachpartition.AdaBoostmethodisappliedtoevaluatetherelativeimportanceofthefeaturesineachpartitionrespectivelyanddistinguishwhetheranobjectisavehicle,asinglehuman,ahumangroup,orabike.Experimentalresultsshowthattheproposedmethodachieveshigherperformanceincomparisonwiththeexistingmethod.

简介：Anobjectmodel-basedtrackingmethodisusefulfortrackingmultipleobjects,butthemaindifficultiesaremodelingobjectsreliablyandtrackingobjectsviamodelsinsuccessiveframes.Aneffectivetrackingmethodusingtheobjectmodelsisproposedtotrackmultipleobjectsinareal-timevisualsurveillancesystem.Firstly,fordetectingobjects,anadaptivekerneldensityestimationmethodisutilized,whichusesanadaptivebandwidthandfeaturescombiningcolourandgradient.Secondly,somemodelsofobjectsarebuiltfordescribingmotion,shapeandcolourfeatures.Then,amatchingmatrixisformedtoanalyzetrackingsituations.Ifobjectsaretrackedunderocclusions,theoptimal'visual'objectisfoundtorepresenttheoccludedobject,andtheposteriorprobabilityofpixelisusedtodeterminewhichpixelisutilizedforupdatingobjectmodels.Extensiveexperimentsshowthatthismethodimprovestheaccuracyandvalidityoftrackingobjectsevenunderocclusionsandisusedinreal-timevisualsurveillancesystems.

简介：ThispaperproposedarobustmethodbasedonthedefinitionofMahalanobisdistancetotrackgroundmovingtarget.Thefeatureandthegeometryofairbornegroundmovingtargettrackingsystemsarestudiedatfirst.Basedonthisfeature,theassignmentrelationoftime-nearbytargetiscalculatedviaMahalanobisdistance,andthenthecorrespondingtransformationformulaisdeduced.Thesimulationresultsshowthecorrectnessandeffectivenessoftheproposedmethod.