搜索
毕业设计（论文）外文文献翻译文献、资料中文题目：远程视频监控系统文献、资料英文题目：文献、资料来源：文献、资料发表（出版）日期：院（部）：专业：电子信息工程班级：姓名：学号：指导教师：翻译日期：2017.02.14外文文献翻译ASystemforRemoteVideoSurveillanceandMonitoringThethrustofCMUresearchundertheDARPAVideoSurveillanceandMonitoring(VSAM)projectiscooperativemulti-sensorsurveillancetosupportbattlefieldawareness.UnderourVSAMIntegratedFeasibilityDemonstration(IFD)contract,wehavedevelopedautomatedvideounderstandingtechnologythatenablesasinglehumanoperatortomonitoractivitiesoveracomplexareausingadistributednetworkofactivevideosensors.Thegoalistoautomaticallycollectanddisseminatereal-timeinformationfromthebattlefieldtoimprovethesituationalawarenessofcommandersandstaff.Othermilitaryandfederallawenforcementapplicationsincludeprovidingperimetersecurityfortroops,monitoringpeacetreatiesorrefugeemovementsfromunmannedairvehicles,providingsecurityforembassiesorairports,andstakingoutsuspecteddrugorterroristhide-outsbycollectingtime-stampedpicturesofeveryoneenteringandexitingthebuilding.Automatedvideosurveillanceisanimportantresearchareainthecommercialsectoraswell.Technologyhasreachedastagewheremountingcamerastocapturevideoimageryischeap,butfindingavailablehumanresourcestositandwatchthatimageryisexpensive.Surveillancecamerasarealreadyprevalentincommercialestablishments,withcameraoutputbeingrecordedtotapesthatareeitherrewrittenperiodicallyorstoredinvideoarchives.Afteracrimeoccurs–astoreisrobbedoracarisstolen–investigatorscangobackafterthefacttoseewhathappened,butofcoursebythenitistoolate.Whatisneedediscontinuous24-hourmonitoringandanalysisofvideosurveillancedatatoalertsecurityofficerstoaburglaryinprogress,ortoasuspiciousindividualloiteringintheparkinglot,whileoptionsarestillopenforavoidingthecrime.Keepingtrackofpeople,vehicles,andtheirinteractionsinanurbanorbattlefieldenvironmentisadifficulttask.TheroleofVSAMvideounderstandingtechnologyinachievingthisgoalistoautomatically“parse”peopleandvehiclesfromrawvideo,determinetheirgeolocations,andinsertthemintodynamicscenevisualization.Wehavedevelopedrobustroutinesfordetectingandtrackingmovingobjects.Detectedobjectsareclassifiedintosemanticcategoriessuchashuman,humangroup,car,andtruckusingshapeandcoloranalysis,andtheselabelsareusedtoimprovetrackingusingtemporalconsistencyconstraints.Furtherclassificationofhumanactivity,suchaswalkingandrunning,hasalsobeenachieved.Geolocationsoflabeledentitiesaredeterminedfromtheirimagecoordinatesusingeitherwide-baselinestereofromtwoormoreoverlappingcameraviews,orintersectionofviewingrayswithaterrainmodelfrommonocularviews.Thesecomputedlocationsfeedintoahigherleveltrackingmodulethattasksmultiplesensorswithvariablepan,tiltandzoomtocooperativelyandcontinuouslytrackanobjectthroughthescene.Allresultingobjecthypothesesfromallsensorsaretransmittedassymbolicdatapacketsbacktoacentraloperatorcontrolunit,wheretheyaredisplayedonagraphicaluserinterfacetogiveabroadoverviewofsceneactivities.Thesetechnologieshavebeendemonstratedthroughaseriesofyearlydemos,usingatestbedsystemdevelopedontheurbancampusofCMU.Detectionofmovingobjectsinvideostreamsisknowntobeasignificant,anddifficult,researchproblem.Asidefromtheintrinsicusefulnessofbeingabletosegmentvideostreamsintomovingandbackgroundcomponents,detectingmovingblobsprovidesafocusofattentionforrecognition,classification,andactivityanalysis,makingtheselaterprocessesmoreefficientsinceonly“moving”pixelsneedbeconsidered.Therearethreeconventionalapproachestomovingobjectdetection:temporaldifferencing;backgroundsubtraction;andopticalflow.Temporaldifferencingisveryadaptivetodynamicenvironments,butgenerallydoesapoorjobofextractingallrelevantfeaturepixels.Backgroundsubtractionprovidesthemostcompletefeaturedata,butisextremelysensitivetodynamicscenechangesduetolightingandextraneousevents.Opticalflowcanbeusedtodetectindependentlymovingobjectsinthepresenceofcameramotion;however,mostopticalflowcomputationmethodsarecomputationallycomplex,andcannotbeappliedtofull-framevideostreamsinreal-timewithoutspecializedhardware.UndertheVSAMprogram,CMUhasdevelopedandimplementedthreemethodsformovingobjectdetectionontheVSAMtestbed.Thefirstisacombinationofadaptivebackgroundsubtractionandthree-framedifferencing.Thishybridalgorithmisveryfast,andsurprisinglyeffective–indeed,itistheprimaryalgorithmusedbythemajorityoftheSPUsintheVSAMsystem.Inaddition,twonewprototypealgorithmshavebeendevelopedtoaddressshortcomingsofthisstandardapproach.First,amechanismformaintainingtemporalobjectlayersisdevelopedtoallowgreaterdisambiguationofmovingobjectsthatstopforawhile,areoccludedbyotherobjects,andthatthenresumemotion.Onelimitationthataffectsboththismethodandthestandardalgorithmisthattheyonlyworkforstaticcameras,orina”stepandstare”modeforpan-tiltcameras.Toovercomethislimitation,asecondextensionhasbeendevelopedtoallowbackgroundsubtractionfromacontinuouslypanningandtiltingcamera.Throughcleveraccumulationofimageevidence,thisalgorithmcanbeimplementedinreal-timeonaconventionalPCplatform.Afourthapproachtomovingobjectdetectionfromamovingairborneplatformhasalsobeendeveloped,underasubcontract