搜索
I摘要随着信息技术的飞速发展，人脸检测是计算机机器视觉的重要研究领域，视频安全监控系统中通常需要针对多个目标的人脸图像进行检测，所以社会发展生产生活离不开多目标人脸检测技术。多目标人脸检测技术在生物检测识别技术迅猛发展的今天，应用尤为广泛。多目标人脸检测是指输入图像内包含多个人脸图像，基于人脸图像进行人脸特征点定位构造几何特征向量进行人脸图像的进一步验证和匹配识别。人脸的几何特征可用于人脸检测，是一种稳定的面部特征，对于多目标的人脸检测技术具有实用的价值和研究意义。本文针对多目标人脸检测方法进行深入研究，基于视频序列图像研究人脸检测，人脸特征点定位，几何特征向量的人脸匹配识别的方法。首先基于视频序列的连续帧图像采用AdaBoost检测算法实现多目标人脸图像的粗定位，包括个数、位置及大小，分析人脸检测算法适用的人脸图像的分辨率、左右旋转角度、上下仰俯角度以及视频序列人脸图像的漏检与误检情况；其次基于灰度统计和面部结构的分布规律进行人脸的特征点定位，包含眼睛、鼻子及嘴等特征点，特征点定位进一步验证人脸检测的准确性；最后根据特征点间连线的距离值构造几何特征向量，特征向量能够全面反映人脸的比例关系以及器官属性，通过计算人脸几何特征向量的加权欧氏距离进行身份识别。采用多目标人脸检测技术在视频图像进行实验，可以达到良好的多目标检测效果。实验表明，多目标人脸检测方法可以有效的对视频序列中不同光照、姿态、表情的人脸图像进行检测，对连续帧图像的多目标人脸图像进行动态分析，有较高的检测率和较好的鲁棒性。关键词：多目标，人脸检测，特征点定位，几何特征向量，相似度判断IIAbstractWiththerapiddevelopmentofinformationtechnology,facedetectionistheimportantresearchfieldofcomputervision,videosecuritymonitoringsystemusuallyneedtodetectmulti-faceimages.Sothedevelopmentofsocietyproductionandlivinginseparablefrommulti-facedetectiontechnology.Themulti-facedetectiontechnologyinthebiometricidentificationtechnologyrapiddevelopmenttoday,iswidelyapplied.Multi-facedetectionreferstocontainmorethanonefaceimagewithintheinputfaceimage,basedonthefacialfeaturepointslocationandgeometriceigenvectorsoffaceimagefurtherverificationrecognitionandmatching.Facialfeatureinformationandgeometricalcharacteristicscanbeusedforfacedetection,isastablefacialfeatures.Thegeometricalcharacteristicsofthehumanfaceformulti-facedetectiontechnologyhasthepracticalvalueandsignificance.Thispapermakesadeepresearchformulti-facedetection,involvedinfacedetectionandfacialfeaturepointslocalizationandtherecognitionofthefacialgeometriceigenvectorsbasedonvideosequence.First,thispaperpresentsthefacedetectionbasedonAdaBoostalgorithmtodetectmanyfaces,includethesizeandlocationinformation.Throughlotsoffaceimagesinvideosequence,itcanfindthatAdaBoostfacedetectionalgorithmiseffectivetodetectthefacialimageswiththeresolution,rotationangleinleftandrightside,pitchingangleandsoon.Aswellasanalysisoffacedetectionalgorithmthemostsuitableoccasion.Secondly,thefeaturepointspositioningbasedongraystatisticandthedistributionofface,realizethefacialeyeandnoseandmouthfeaturelocation.Facialfeaturepointslocationresultsdirectlyverifytheaccuracyoffacedetection.Finally,thegeometriceigenvectorscomposedofthedistancebetweenthefacialfeaturepoints,itcancomprehensivereflectthefacialproportionrelationshipbetweenorgansandproperties.BycalculatingweightedEuclideandistancebetweenthefacialgeometriceigenvectors,itisusedtoidentifythemulti-faceimagesofvideosequence.Toexperimentonvideoimagewithmulti-facedetectionmethoddiscussedabove,achievedagoodmulti-facedetection.Extensiveexperiencesandanalysisillustratesthatthesemethodscanbeapplieddifferentexpressionsandrotationangleandsizeoffacedetection,realizethedynamicanalysisofsuccessiveframesmulti-faceimage.Multi-facedetectioncanIIIbeeffectivelyrealizedfacedetectionwithhighdetectionrate,speed,strongrobustness.KeyWords:Multi-objective,Facedetection,Facialfeaturepointslocalization,Geometriceigenvectors,SimilarityjudgmentIV目录摘要...........................................................................................................................................IAbstract....................................................................................................................................II第1章绪论............................................................................................................................11.1人脸检测方法研究的目的及意义...........................................................................11.2多目标人脸检测方法研究动态...............................................................................21.2.1多目标人脸检测的研究现状........................................................................21.2.2多目标人脸检测研究内容............................................................................41.3本文工作概述和结构...............................................................................................5第2章基于AdaBoost人脸检测方法研究..........................................................................72.1AdaBoost人脸检测算法原理...................................................................................72.1.1积分图与矩形特征........................................................................................82.1.2分类器的设计..............................................................................................102.2人脸检测算法分析.................................................................................................112.2.1人脸检测算法适用条件..............................................................................112.2.2视频序列人脸检测算法漏检与误检分析..................................................142.3基于AdaBoost算法人脸图像的粗定位...............................................................152.4本章小结.................................................................................................................19第3章人脸特征点定位方法研究......................................................................................203.1人脸特征点的选取............................................................................