人臉頭部姿態(tài)(龍1)

人臉頭部姿態(tài)2015-2Watson人臉頭部姿態(tài)的跟蹤的連接。/display/VHTK/Watson#Watson-ConfigurationforStereoCamera//vision/watson/MultiSenseMultiSenseisaperceptionframeworkthatenablesmultiplesensingandunderstandingmodulestointer-operatesimultaneously,broadcastingdatathroughthePerceptionMarkupLanguage.MultiSensecurrentlycontainsGAVAM,CLMFaceTrackerandFAASTwhichcanbeusedwithawebcamorKinect.TheToolkitprovidesanexampleofhowtousetheMultiSenseframework(alsoknownasmultimodalframework,developedbyMulticompLab).多傳感器是一個認知框架，使多個感知和理解模塊進行相互操作的同時，通過感知的標記語言廣播數(shù)據(jù)。多傳感器目前包含gavam，CLM

facetracker和FAAST，它們可以使用一個攝像頭或Kinect。該工具包提供了一個例子，如何利用多傳感器框架（也被稱為多模態(tài)框架，由multicomp實驗室開發(fā)）。MultiSense-GAVAMFaceTrackerprobabilisticframework:GeneralizedAdaptiveView-basedAppearanceModel(GAVAM)Accuratelyestimatingtheperson’sheadpositionandorientationintegratestheadvantagesfromthreeoftheseapproaches:(1)theautomaticinitializationandstabilityofstaticheadposeestimation,(2)therelativeprecisionanduser-independenceofdifferentialregistration,and(3)therobustnessandboundeddriftofkeyframetrackingGAVAMmodelcanbeusedtoestimateheadpositionandorientationinreal-timeusingasimplemonocularcamera./?s=gavam&x=0&y=0概率框架：廣義的自適應的基于視圖的外觀模型（gavam）準確估計的人的頭部的位置和方向結合了三種方法先進的地方：（1）自動初始化和靜態(tài)頭部姿態(tài)估計的穩(wěn)定性，（2）相對精度和不同用戶的獨立性（3）關鍵幀跟蹤的魯棒性和有界漂移gavam模型可以使用一個簡單的單眼相機實時估計頭的位置和方向。MultiSense-CLM3DConstrainedLocalModel(CLM-Z)robustfacialfeaturetrackingundervaryingposecombinearigidheadposetrackerwithCLM-Zegratesbothdepthandintensityinformationinacommonframework3DConstrainedLocalModelforRigidandNon-RigidFacialTracking,TadasBaltru?aitis,PeterRobinson,Louis-PhilippeMorency.IEEEComputerVisionandPatternRecognition2012http://www.cl.cam.ac.uk/~tb346/res/clm-z.html三維約束的局部模型（clm-z）不同人臉姿態(tài)特征點跟蹤的魯棒性結合CLM—Z的剛性頭部姿態(tài)跟蹤有利于剛性頭部跟蹤集成深度和強度信息在一個共同的框架三維約束的局部模型的剛性和非剛性的面部跟蹤，TadasBaltru?aitis,PeterRobinson,Louis-PhilippeMorency.IEEEComputerVisionandPatternRecognition2012MultiSense-FAASTFlexibleActionandArticulatedSkeletonToolkit(FAAST)FAASTismiddlewaretofacilitateintegrationoffull-bodycontrolwithgamesandVRapplicationsusingeitherOpenNIortheMicrosoftKinectforWindowsskeletontrackingsoftware.FAASTincludesacustomVRPNservertostreamuptofouruserskeletonsoveranetwork,allowingVRapplicationstoreadtheskeletaljointsastrackersusinganyVRPNclient.Additionally,thetoolkitcanalsoemulatekeyboardinputtriggeredbybodypostureandspecificgestures.Thisallowstheuseraddcustombody-basedcontrolmechanismstoexistingoff-the-shelfgamesthatdonotprovideofficialsupportfordepthsensors./mxr/faast/E.Suma,D.Krum,B.Lange,S.Koenig,A.Rizzo,andM.Bolas.Adaptinguserinterfacesforgesturalinteractionwiththe?exibleactionandarticulatedskeletontoolkit.Computers&Graphics,37(3):193–201,2013.靈活的動作骨架工具包（FAAST）FAAST是中間件便于集成的全車身控制游戲和虛擬現(xiàn)實應用，使用OpenNI或者微軟Kinect骨骼跟蹤軟件。FAAST包括自定義演示服務器流高達四用戶的骨骼在網(wǎng)絡虛擬現(xiàn)實應用，允許讀取骨骼關節(jié)使用任何演示客戶跟蹤。此外，該工具箱還可以模擬鍵盤輸入的身體姿勢和手勢觸發(fā)特定。這允許用戶添加自定義的控制機制身體到現(xiàn)有的游戲，不提供官方支持的深度傳感器。/mxr/faast/E.Suma,D.Krum,B.Lange,S.Koenig,A.Rizzo,andM.Bolas.Adaptinguserinterfacesforgesturalinteractionwiththe?exibleactionandarticulatedskeletontoolkit.Computers&Graphics,37(3):193–201,2013FAAST（FlexibleActionandArticulatedSkeletonToolkit）是一個中間件類庫用來連接Kinect的手勢交互界面和傳統(tǒng)的人機交互界面。這個類庫由南加利福利亞大學的創(chuàng)新技術研究院編寫和維護。最初FAAST是使用OpenNI結合Kinect來編寫的一個手勢識別類庫。這個類庫真正優(yōu)秀的地方在于它能夠?qū)⒋蠖鄶?shù)內(nèi)建手勢映射到其他API中，甚至能夠?qū)⑹謩萦成涞芥I盤點擊中來。這使得一些Kinect技術愛好者(hacker)能夠使用這一工具來使用Kinect玩一些視頻游戲，包括一些例如《使命召喚》(CallofDuty)的第一人稱射擊游戲，以及一些像《第二人生》(SecondLife)和《魔獸世界》(WorldofWarcraft)這樣的網(wǎng)絡游戲。目前FAAST只有針對OpenNI版本，但是據(jù)最新的消息，F(xiàn)AAST現(xiàn)在正在針對KienctSDK進行開發(fā)。MultiSense-ssiLibraryTheSocialSignalInterpretation(SSI)frameworkofferstoolstorecord,analyseandrecognizehumanbehaviorinreal-time,suchasgestures,mimics,headnods,andemotionalspeech.Itsupportsstreamingfrommultiplesensorsandincludesmechanismsfortheirsynchronization.InparticularlySSIsupportsthemachinelearningpipelineinitsfulllengthandoffersagraphicalinterfacethatassistsausertocollectowntrainingcorporaandobtainpersonalizedmodels.Italsosuitsthefusionofmultimodalinformationatdifferentstagesincludingearlyandlatefusion.SSIiswritteninC++andsourcecodeisavailableunderLGPL.rmatik.uni-augsburg.de/lehrstuehle/