用行人轮廓的分布直方图分类和识别步态

《用行人轮廓的分布直方图分类和识别步态》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、万方数据if．算机研究与发展ISSN1000—12391CN11-1777／TPJournalof【’omputerRcsearchandDevclopmcnt46(2)：295—301，2009用行人轮廓的分布直方图分类和识别步态陈实1一马天骏2高有行21(浙江万里学院设计学院浙江宁波315100)2(西安电子科技大学计算机学院西安710071)(schen@mail．xidian．edu．cn)GaitRecognitionUsingDistributionsofSilhouetteFeatureChenShil～，MaTianjun2，andGaoYouxin921(Faculty

2、ofDesign，ZhejlankWanliUniversity，Ningbo，Zhejiang315100)2(SchoolofComputerScienceandTechnology。XidianUniversity，Xi’an710071)AbstractVision—basedhumanidentificationatadistanceinsurveillancehasrecentlygainedmoreattentions．Gaithastheadvantagesofbeingnon—invasiveanddifficulttoconceal，andisalsotheonlyp

3、erceivablebiometricatadistance．Thispaperintroducesanovelfeaturerepresentatiorimethodforgaitanalysisandrecognitionapplications．Themethodincludesfollowingsteps：first，silhouetteextractionisperformedforeachimagesequence．Secondly，thedistributionsofsampledpointsftomthehumanlocalsilhouetteareanalyzedand

4、thegaitcycleisdetectedbyahistogram—basedapproach．Thirdly．bytilinganddispersingallimageframesacrossonegaitcycleinatwo-dimensionalplanealongaringframebyframeatafixedinterval，acontextualstancesappearancemodelisbuilt．Thegaitappearancemodelconsistsofthestructuralinformationoftheindividualsilhouetteand

5、contextualsilhouettescenteredatthecurrentframeinthepolar—plane．Withadesignedinvarianthistogram—baseddescriptor，thegaitappearancecharacteristicsaredescribedasasequenceofshapedistributions．ThesedistributionsarefinallyusedtoachievegaitrecognitionbasedonJeffreydivergencematchingcriterionanddynamictim

6、ewarpingtechnology．Recognitioncapabilityisillustratedbyan87．59％CCRonSotondatabaseandtheresultshowsthatourapproachoutperformsexistingmethods．Keywordsgaitrecognition；behavioralbiometrics；gaitanalysis；humansi[houette；shapematching；distributions摘要现有的步态识剐方法对行人轮廓匹配的鲁棒性差，识别率不高．提出了一种基于轮廓直方图分布的行人步态识别方法．首先

7、提取行人二值轮廓序列；然后通过人体局部轮廓的点分布直方图获取步态周期；继而构造表达帧问关系的周期步态平面，设计一种直方图分布的描述子获得帧姿态特征值，计算出姿态轮廓特征分布间的Jeffcry距离，结合动态时间规整技术获取了测试序列和参考序列间的匹配相似度，最终完成了识别．在Soton步态数据库上进行了实验，提出算法的正确识别率可达87．59％，与相关文献的对比分析表明算法是有效的．关键词步态识别；行为特征；步态分析；行人轮廓；形状匹