基于视频序列的运动目标检测与跟踪技术研究

《基于视频序列的运动目标检测与跟踪技术研究》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、西南交通大学硕士学位论文基于视频序列的运动目标检测与跟踪技术研究姓名：朱宏申请学位级别：硕士专业：信号与信息处理指导教师：王建英20080301西南交通大学硕士研究生学位论文第1工页AbstractThetechnologyofvideobasedonobjectdetectionandtrackingisoneofthehotspotsinthefieldofcomputervision，whichiSalsothebasicintheapplicationsofsmartsurveillance，human

2、．machineinterface．mobilerobotsnavigation,industrialrobotshand—eyesystemandSOon．Inreallife。movingtargetdetectionandtrackingextensivelyinvolvesinhumantrackingandidentification．intelligenttransport，trafficflowmonitoring，etc．Duringthepastfewyears，muchefforthasbe

3、enmadetoenhancetheperfc)rmanceofvisualtrackingbycombiningmultiplefeatures．Inthispaper,afastalgorithmoftargetdetectioninstaticbackgroundvideosequencesandatrackingalgorithmisproposedbasedonregioncorrelationdescriptor,whichcanprovideanelegantsolutiontofusemulti

4、plefeaturesforrepresentingtheobject．Besides，wesuggestamethodtofindthebestmatchingregionbycalculatinga“centroid”estimatedfromasetofcandidatepointsinsearchingwindow．Maincontentcanbesummarizedasfollows：1．Inviewofthefixedvideosequence．arapidmethodofmovingtargetd

5、etectionalgorithmisproposed．Duetothemovingtargetappearsindifferentlocationsineachframe。itCallbeinnovativelyviewedasrandomnoise．Therefore。thebackgroundmodelCanbeex仃actedquicklythroughthemethodofthemeansoflimitedframes．Then,thispaperpresentshigh-bit—layerexclu

6、siveOR(XOR)algorithmtoextractmovingtarget．Comparetotraditionalbackgroundsubtractionalgorithm，newalgorithmavoidtheinfluenceofcamerajitterandthesunshine．Theeffectiveandaccuracyofnewalgorithmhasbeendemonstratedbyexperiments．2．Consideringtheinsu伍cientofextractin

7、gfeatureonlyfromthegraylevelimagetraditionally，thispaperpresentsaregioncorrelationdescriptortofusemultiplefeaturestorepresenttheobject．Firstly，thealgorithmestablishesatemplate，includingtheinformationofcolor,textureandedgegradient．SecondlyForstnerdistanceisus

8、edasacriteriatomeasurethematchingextentbetweentemplateandthesearchingarea．Theexperimentalresultsshowthatthetrackingemciencyhasbeenenhancedina1argescale．3．Inordertoreducethecomputation．searchinga