资源描述:
《尺度自适应扩展mean shift跟踪算法》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、AnextendedMeanshifttrackingalgorithmwithscaleadaptiveThesisSubmittedtoNanjingUniversityofPostsandTelecommunicationsfortheDegreeofMasterofEngineeringByXiejieSupervisor:Prof.BaoyunWangFebruary2013IIIII摘要随着人工智能、数字图像处理、模式识别等领域的突飞猛进的发展,智能视频监控日益成为一门应用广泛的综合性学科。其中目标跟踪技术是一种识别目标,提取位置信息,自动跟踪目标运动的技术
2、,是该领域中一个具有重要研究意义的课题。论文的主要内容如下:1.本文为了准确的表述目标物体特征,引入了一种新型纹理特征表示法LBP纹理,并8,1将这种纹理和颜色特征结合起来,用直方图的形式共同描述目标模型,使得目标表示更精确。实验结果证明,该方法能更明显的反映出目标特性,而算法复杂度没有提高太多。2.基于传统Meanshift跟踪算法易受到遮挡、背景信息干扰等情况。本文采用一种扩展的Meanshift算法,它是将纹理和背景权重相结合,采用LBP中五种特征来表示8,1目标纹理,再引入背景加权因子,使得跟踪精度提高明显。通过试验证明该种方法对于目标跟踪有很好的效果。3.同时
3、为了改善跟踪算法中尺度自适应,本文提出了通过相似度比较更新带宽的新方法。分别计算当前帧和前一帧中中心像素点、边缘像素的相似度函数,,通过1212比较两者大小判断出目标尺度变化情况,以实现尺度更新。实验证明该尺度更新法原始尺度更新法更精确。关键词:像素点,目标跟踪,Meanshift,纹理,直方图,尺度自适应IVAbstractAstheincreasinglydevelopmentofartificialintelligence,digitalimageprocessing,patternrecognitionandotherareas,intelligentvi
4、deomonitoringisincreasinglybecomingawideapplicationofcomprehensivediscipline.Andinthisfield,thetargettrackingtechnologyisakindofenergyfromtheimagesignalreal-timeidentificationoftarget,extracttargetlocationinformation,automatictrackingtargetmotiontechnology.Soitisanimportantresearchsignif
5、icanceofthesubject.Atpresenttherearemanymethodsinthetargettrackingtechnology,basedonthepreviousresearchs,IputforwardanadaptiveMeanshifttrackingalgorithm..Inthispaper,it’sintroducestheprobabilitydensityhistogramwhichbasedontheR,G,Bcolorandtextureoftheobject,andbringinanewtypeoftexturefeat
6、ureisLBPtexturefeature.Aim8,1totargetmodeldescription,combinethetextureandcolorfeaturestogether,onlytojointhetexturefeatureandthecolorfeaturessimply,thecomputationalcomplexityofthismethodisveryhigh,Soitisnotsuitablefortheneedofdailyapplication.Therefore,thisuseofthefivekindsoftexturefeat
7、uremodeltorepresentobjectstexture.Theexperimentalresultsshowthatthismethodcansignificantlymorereflectthecharacteristicsoftarget,andalgorithmcomplexitydoesn'timprovetoomuch.SecondlybasedonthetraditionalMeanShifttrackingalgorithmissusceptibletobesheltered,backgroundinformat