简介:MotionEstimationBasedonGlobalandLocaCompensabilityAnalysisTXMotionEstimationBasedonGlobalandLocalCompensabilityAnalysisFanJin...
简介:Anovelimagestabilizationsystemispresented,whichconsistsofaglobalfeaturepointtrackingbasedmotionestimation,aKalmanfilteringbasedmotionsmoothingandanimagemosaicbasedpanoramiccompensation.Theglobalmotionisestimatedusingfeaturepointmatchinganditerationwiththeleast-squaremethod.Then,theKalmanfilterisappliedtosmooththeoriginalmotionvectorstoeffectivelyalleviateunwantedcameravibrationsandfollowtheintentionalcamerascan.Lastly,thelossinformationofim...
简介:Theproblemofrobustifyinglinearquadraticregulators(LQRs)foraclassofuncertainaffinenonlinearsystemsisconsidered.First,theexactlinearizationtechniqueisusedtotransformanuncertainnonlinearsystemintoalinearoneandanoptimalLQRisdesignedforthecorrespondingnominalsystem.Then,basedontheintegralslidingmode,adesignapproachtorobustifyingtheoptimalregulatorisstudied.Asaresult,thesystemexhibitsglobalrobustnesstouncertaintiesandtheidealslidingmodedynamicsisthesameasthatoftheoptimalLQRforthenominalsystem.Aglobalrobustoptimalslidingmodecontrol(GROSMC)isrealized.Finally,anumericalsimulationisdemonstratedtoshowtheeffectivenessandsuperiorityoftheproposedalgorithmcomparedwiththeconventionaloptimalLQR.
简介:BlockmatchinghasbeenusedformotionestimationandmotioncompensationintheAu-dioandVideoStandard(AVS)foryears.Whilehavinganacceptableperformanceindescribingmo-tionbetweenframes,itrequiresquiteafewbitstorepresentthemotionvectors.Incertaincircum-stances,theuseofglobalmotionestimationandcompensationwouldperformequallywellorevenbetterthantheblockmatchingintermsofmotionaccuracy,whileitresultsinthecodingofglobalmotionmodelparameters.Inthisletter,wemodifyanAVScoderbyadding(1)sixglobalmotionmodelparameterstotheframeheader,and(2)modeselectionamongINTRA,SKIP,INTER-16×16,INTER-16×8,INTER-8×16,INTER-8×8,andGlobalMotionCompensation(GMC)modesbyLa-grangeoptimalrate-distortioncriteria.Simulationresultsdemonstratethatover0.1dBimprovementinPSNRisobtainedcomparedtotheAVScoderforanaveragecodedP-framewiththesamebit-rate.
简介:Inthispaper,wepresentanovelregion-basedactivecontourmodelbasedonglobalin-tensityfittingenergyinavariationallevelsetframework.Meanwhile,aninternalenergytermisin-troduced,anditforcesthelevelsetfunctiontobeclosetoasigneddistancefunction.Imageglobalinformationutilizedefficientlymakestheproposedmodelinsensitivetonoise,andtheintroducedpenaltytermcanavoidthecostlyre-initializationfortheevolvinglevelsetfunction,whichnotonlyspeedsupthecontourevolvement,butalsoimprovesaccuracyofthefinalcontour.Comparisonswithotherclassicalregion-basedmodels,suchasChan-VesemodelandRegion-ScalableFitting(RSF)model,showtheadvantagesofourmodelintermsofefficiencyandaccuracy.Moreover,themodelisrobusttonoise.
简介:Inthispaper,anovelframework,namedasglobal-localfeatureattentionnetworkwithrerankingstrategy(GLAN-RS),ispresentedforimagecaptioningtask.Ratherthanonlyadoptingunitaryvisualinformationintheclassicalmodels,GLAN-RSexplorestheattentionmechanismtocapturelocalconvolutionalsalientimagemaps.Furthermore,weadoptrerankingstrategytoadjustthepriorityofthecandidatecaptionsandselectthebestone.TheproposedmodelisverifiedusingtheMicrosoftCommonObjectsinContext(MSCOCO)benchmarkdatasetacrosssevenstandardevaluationmetrics.ExperimentalresultsshowthatGLAN-RSsignificantlyoutperformsthestate-of-the-artapproaches,suchasmultimodalrecurrentneuralnetwork(MRNN)andGoogleNIC,whichgetsanimprovementof20%intermsofBLEU4scoreand13pointsintermsofCIDERscore.
简介:Somesufficientconditionsfortheglobalexponentialstabilityandlowerboundsontherateofexponentialconvergenceofthecellularneuralnetworkswithdelay(DCNNs)areobtainedbymeansofamethodbasedondelaydifferentialinequality.Themethod,whichdoesnotmakeuseofanyLyapunovfunctionals,issimpleandvalidforthestabilityanalysisofneuralnetworkswithdelay.Somepreviouslyestablishedresultsinthispaperareshowntobespecialcassesofthepresentedresult.
简介:CooperativepathdynamicplanningofaUCAV(unmannedcombatairvehicle)teamnotonlyconsidersthecapabilityoftaskrequirementofsingleUCAV,butalsoconsidersthecooperativedynamicconnectionamongmembersoftheUCAVteam.AcooperativepathdynamicplanningmodeloftheUCAVteambyapplyingaglobaloptimizationmethodisdiscussedinthispaperandthecorrespondingmodelisbuiltandanalyzed.Bytheexamplesimulation,thereasonableresultacquiredindicatesthatthemodelcouldmeetdynamicplanningdemandunderthecircumstanceofmembershipfunctions.Themodeliseasytoberealizedandhasgoodpracticability.
简介: