资源描述：
《人运动的视觉分析综述》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。
1、第25卷第3期计算机学报Vol.25No.32002年3月CHINESEJ.COMPUTERSMar.2002人运动的视觉分析综述王亮胡卫明谭铁牛(中国科学院自动化研究所模式识别国家重点实验室北京100080)摘要目前,人运动的视觉分析是计算机视觉领域中最活跃的研究主题之一,其核心是利用计算机视觉技术从图像序列中检测、跟踪、识别人并对其行为进行理解与描述,它在虚拟现实、视觉监控、感知接口等领域均有着广阔的应用前景.人运动的视觉分析系统一般遵从下述的处理过程:(1)运动检测,(2)运动目标分类,(3)人的跟踪,(4)行为理解与描述.该文将重点从此四个方面回顾人运
2、动分析的发展水平和常用的处理方法,并对研究难点及未来的发展趋势作了较为详细的分析.关键词人的运动,视觉分析,运动检测,跟踪,行为理解中图法分类号:TP391ASurveyofVisualAnalysisofHumanMotionWANGLiangHUWei-MingTANTie-Niu(NationalLaboratoryofPatternRecognition,InstituteofAutomation,ChineseAcademyofSciences,Beijing100080)AbstractVisualanalysisofhumanmotioniscu
3、rrentlyoneofthemostactiveresearchtopicsinthedomainofcomputervision.Thisstronginterestisdrivenbyawidespectrumofpromisingap-plicationsinmanyareassuchasvirtualreality,smartsurveillance,perceptualuserinterface,content-basedimagestorageandretrieval,athleticperformanceanalysis,etc.Humanmo
4、tionanalysisaimsatattemptingtodetect,trackandidentifypeople,andmoregenerally,tounder-standhumanbehaviors,fromimagesequencesinvolvinghumans.Thispaperprovidesacompre-hensivesurveyofrecentdevelopmentsofvision-basedhumanmotionanalysis,anditkeepsupwiththelatestresearchrangingmainlyfrom19
5、95to2000.Differentfrompreviousreviews,ouremphasisisonfourmajorissuesinvolvedinageneralframeworkofhumanmotionanalysis,namelymotiondetection,movingobjectclassification,humantracking,andactivityrecognitionanddescription.Thispaperfocusesmoreonoverallmethodsandgeneralcharacteristicsinvol
6、vedintheabovefourissues,soeachissueisaccordinglydividedintosub-processesandcategoriesofapproachessoastoprovidemoredetaileddiscussions.Atfirst,weintroducesomepotentialap-plicationsofhumanmotionanalysis.Then,variousexistingmethodsforeachkeyissueareclear-lydiscussedtoexaminethestate-of
7、-the-artinhumanmotionanalysis.Motiondetectionprovidesafocusofattentionforlaterprocessesbecauseonlythosechangingpixelsneedbeconsidered.Threetypesoftechniquesareaddressed,namelybackgroundsubtraction,temporaldifferencingandopticalflow.Asfarasmovingobjectclassificationisconcerned,shape-
8、basedormotion-based
