网络流媒体内容分析关键技术研究

上海交通大学硕士学位论文网络流媒体内容分析关键技术研究姓名：陈枫申请学位级别：硕士专业：通信与信息系统指导教师：杨树堂20051201I“”IIMarkovHMMF--FSMFSMMean-ShiftIFGTMarkovIIIRESEARCHONKEYTECHNOLOGYINNETWORKSTREAMINGMEDIACONTENTANALYSISABSTRACTWiththerapiddevelopmentofcomputerhardwaretechnologyandnetworktechnology,peoplenotonlywanttobrowsewebsanddownloadfiles,butalsowanttoappreciatereal-timevideoandaudio.Therefore,streamingmediatechnologyemergedandrapidlydeveloped.StreamingmediaisconsideredtobethemainstreamofInternetapplicationinthefuture.Andnow,ithasbeenappliedinmulti-medianews,demonstratingthroughinternet,on-lineliving-broadcast,netadvertisement,e-business,video-on-demand,remoteeducationandetc.ItrevolutionizesentireInternetarchitecture.Inaddition,ourlifeisaffectedbythespreadofstreamingmediainformation.Buteverycoinhastwosides.Streamingmediabringstheill-healthyinformationtopeoplewhichwouldinfluenceonthestabilityofoursocietyandpeople’sideology.Thus,weurgentlyneedanewcontentanalysistechnologytoprotectsecurityofnetworkstreamingmedia.IVOntheotherhand,withsignificantdevelopmentofthetechnologyinthefieldofvideocapture,compression,transmissionandstorage,content-basedvideoanalysishasbecomeahottopicinfieldofcomputervision.Thethreelevelofcontentanalysistechnologyconsistofvideosegmentation,videoobjecttrackingandactioncomprehensionofvideoobject.Amongdifferentareasofcontent-basedanalysis,videoobjecttrackingisanimportantresearchorientationthatfullofgreatchallenges.TheapplicationprospectsofvideoObjecttrackinginvideoprocessing,videocompressionandcomputervisionissowidethatithasreceivedenoughattentionintheacademicsocietyandmanytheoriesandapproacheshavebeenpresentedintheliteratures.Weanalyzethecontent-basedanalysistechnologyofstreamingmediafocusingontechnologyofvideoobjecttracking.Andafterdescribingbasicconceptsofstreamingmediaapplicationsystem,anovelvideoobjecttrackingalgorithmsispresentedinthispaper.Thebasicideaofthealgorithmcanbebrieflydescribedasfollow:Generally,theregion-basedtrackinginatemporalimagesequencecanbedescribedasasegmentationofcurrentframeintoasetofnon-overlappingregions:thetrackingregionsandthenon-trackingregion.ThesegmentationisviewedtobeaMarkovlabelingprocess.Wefurtherpresentaprobabilisticframeworkfortrackingobjectsintheformoflocaldynamicsegmentation.WeviewthesegmentationtobeaMarkovVlabelingprocessandabstractitasaMAPproblem.IntheBayesianformulation,weexploittheFeature-Spatial-Measuredistributionoflocalareaastheconditionaldistribution.TheFeature-SpatialvectorisusedtoconstraintheappearanceofregionwhiletheMeasurevectorisusedtoconstrainthelabelofthepixelsintheregion.OnedriveforcetotheintroductionofFSMdistributionisstilltheHMMFmodelthatmakesitpossibletoestimatetheMeasurefieldbytheminimizationofadifferentiablefunction.Mean-shiftprocedureandIFGTtechniqueareusedtofurtheralleviatethecomputationalcosts.Experimentalresultsshowthatthisalgorithmcanachievereal-timetrackinginsomeoccupations.Atlast,weappliedthismethodintothesystemnamedInformationdetectionandContentanalysisforNet-Basedstreamingmedia.Andtheresulthasprovedourtrackerworkedwell.KEYWORDSnetworkstreamingmedia,contentanalysis,ObjectTracking,StatisticalModel,HiddenMarkovRandomField2006111InternetIPVideo-AudioStream14.4kbps10MbpsYahoo,1.11.1.1StreamingMediaInternet/Intranet256K1”1.1.2:1)4512)/3)rtsp1.21.2.1//2001RealNetworks32003431.626.6%PioneerConsultingLLC20002.43Gbit/s2003170Gbit/s20086400Gbit/sQoS(3G)(Live)1.2.24R()VisionNEXTeefindMimesweeperpornsweeperMargaretFleckIBMQBICAttrasoftImageFinderMWLabsImatchLookThatUpImage-FilterLookThatUpServerPartnerServer,1)5(1),[1-6]GMMKKDE[7](2)[8]MAPGibbs(3)Dalembier[9](a)(b)(c)watershed(d)[10]Watershed(4)SnakeKass[11]SnakeParametricGeometric()Snake6Parametric2)[Gregory1998](1)Euclidean[12],(2)[1314][15,16]7[17](3)[18]Mansouri(RegionCompetition)[19]3)1.38--92.1()Internet/IntranetPCInternet(RTSP)A/VCDRTSPA/VRTP/UDPA/VA/VRTP/UDPRTSP/TCPA/V2-110WebBrowserWebServerA/VServerPlayerFileHTTP/TCPDataTCP/UDPCommandRTSP/TCP(UDP)2-1Figure2-1BasicPrincipleofStreamingTransport(RealtimeStreaming)(ProgressiveStreaming)RTSPHTTP2.1.1HTTPHTTPHTTPFTP2.1.2HTTP11QuickTimeStreamingServerRealServerWindowsMediaServerHTTPRTSP(RealTimeStreamingProtocol)MMS(MicrosoftMediaServer)2.2InternetRTPRTCPRTSPRSVPTCP/IP2.2.1RTPRTP(RealtimeTransportProtocol)InternetIETF(InternetEngineeringTaskForce)RTPUDP(UDP)UDP(Multicast)(Unicast),RTPTCPATMRTPRTPRTCPRTPRTCPRTPRTCP(RealtimeTransportControlProtocol)RTPRTPRTCPRTCP12RTPRTCPRTPRTPRTCPRTPRTP2-212RTP2-2RTPFigure2-2RTPPacketHeader2.2.2RTCPRTCPUDPRTCPRTPIPIPRTCPRTCP(CNAME)RTPSSRCCNAMECNAMERTPRTCPRTP13RTCPIPRTP3RTPUDPRTPRTCPRTCPRTPRTCPRTCPRTPRTCP2-3PlayerA/VserverCommandRTCP/RTP(UDP)DataRTP/UDP2-3RTPRTCPFigure2-3CooperationofRTPandRTCP2.2.3RTSPRTSP(RealTimeStreamingProtocol)RealNetworksNetscapeIPRTSPRTPRTCPTCPRTPHTTPRTSPHTTPHTMLRTPHTTPRTSPRTSP(RTSP)RTSPUDPUDPTCPRTP(RTSP)14RTSPRTSPTCPRTSPRTSPRTSPUDPRTSPRTPRTSPHTTP/1.1HTTPRTSPRTSPHTTPHTTPRTSPHTTPRTSPRTSPHTTPRTSPHTTPHTTPRTSPHTTPHTTPRTPRTSPRTPRTSPRTSPTCPRTSPUDPUDPTCPRTSP554