外文翻译---一个鲁棒的基于机器视觉运动目标检测与跟踪系统1111.docx

外文翻译---一个鲁棒的基于机器视觉运动目标检测与跟踪系统1111.docx

《外文翻译---一个鲁棒的基于机器视觉运动目标检测与跟踪系统1111.docx》由会员分享，可在线阅读，更多相关《外文翻译---一个鲁棒的基于机器视觉运动目标检测与跟踪系统1111.docx（14页珍藏版）》请在冰豆网上搜索。

附录1外文文献翻译

（1）原文：

A RobustVision-basedMovingTargetDetectionandTrackingSystem

Abstract

111thispaperwepresentanewalgoritlunforreal~timedetectionandtrackingofmovingtargetsinterrestiialscenesusingamobilecamera.Ouralgo1itlunconsistsoftwomodes:

detectionandtracking.In小edetectionmode,backgi·oundmotionisestimatedandcompensatedusinganaffinetransfonnation.TI1eresi1ltantmotionrectifiedimageisusedfordetectionoftl1etargetlocationusingsplitandmergealgo1itlun.Wealsocheckedotherfea血esforprecisedetectionofthetargetlocation.Whenthetargetisidentified,algo1itlun郅itchesto小etrackingmode.ModifiedMoravecoperatorisappliedtotl1etargettoidentifyfeaturepoints.TI1efea血epointsarematchedwitl1pointsintheregionofinterestinthecmrentframe.九ecorrespondingpointsarefurtl1errefinedusingclispa1ityvectors.九eh·ackingsystemiscapableoftargetshaperecove1yandtl1ereforeitcansi1ccessfullytracktargetswitl1va1yingdistancefromcameraorwhiletl1ecameraiszooming.Localandregionalcomputationshavemade小ealgoritlunsuitableforreal-timeapplications.TI1erefinedpointsdefinetl1enewpositionoftl1etargetintl1ecurrentframe.Expe1imentalresi1ltshave

showntliattl1ealgo1itlunisreliableandcansucces函llydetectand订acktargetsinmostcases.

Key words:

realtimemovingtargeth·ackinganddetection,featurematching,affinetransfonnation,vehicletracking,mobilecameraimage

1Introduction

Vistialdetectionandtrackingisoneoftl1emostchallengingisst1esincomputervision.Applicationofthevist1aldetectionand订ackingarenumerousandtheyspanawiderangeofapplicationsincludingstuveillancesystem,vehiclett·ackingandaerospaceapplication,tonameafew.Detectionandtt·ackingofabstracttargets（e.g.

vehiclesingeneral）isave1ycomplexproblemanddemandssoplristicatedsolutionsusingconventionalpatternrecogtritionandmotionestimationmethods.Motion-basedsegtnentationisoneoftl1epowe1fultoolsfordetectionandtt·ackingofmovingtargets.It issimpletodetectmovingobjectsinimageseqt1encesobtainedbystationa1ycamera （1],（2],小econventional difference-basedmetl1odsfailtodetectmovingtargetswhenthecameraisalsomoving.111the caseofmobilecamera alloftheobjectsintl1eimagesequencehaveanapparentmotion,winchisrelatedto小ecameramotion.Anumberofmetl1odshavebeenproposedfordetectionoftl1emovingtargetsinmobilecameraincludingdirectcameramotionparametersestimation（3）,opticalflow（4）,（5）,andgeometrictransfonnation（6）,（7）.Directmeast1rementofcameramotionparametersistl1ebestmetl1odfor cancellation oftheapparei1tbackgt·oundmotionbutinsomeapplicationitisnotpossibletomeast1retheseparametersdirectly.Geomett·ictt·ansfonnationmethodshavelowcomputationcostandaresuitablefor

realtimepurpose.111thesemetl1od,samrifonnbackgt·otmdmotionisasstuned.知

affinemotionmodelcouldbeusedtomodeltlrismotion.Whentheapparentmotionoftl1ebackgt·01111disestimated,itcanbeexploitedtolocatemovingobjects.111tlrispaperweproposeanewmethodfordetectionandtt·ackingofmovingtargetsusingamobilemonocularcamera.Ouralgo1itl11nhastwomodes:

detectionandtracking.Tirispaperisorga1rizedasfollows.111Section2,小edetectionproce邮eis由scussed.Section3desct·ibesthett·ackingmethod.Experimentalrest1ltsareshowt1inSection4andconclusionappearsinSection5

2Targetdetection

111小edetectionmodeweusedaffinetransfonnationandLMeclS（Leastmedian叩arecl）metl1odforrobustestimationof小eapparentbackgroundmotion.Aftertl1ecompensationoftl1ebackgi·oundmotion,weapplysplitandmergealgorithmtothedifferenceofcurrei1tframeandtl1e订ansfonnedpreviousframetoobtainanestimationoftl1etargetpositions.Ifnotargetisfound,小enitmeanseitl1ertl1ereisnomovingtargetinthesceneor,therelativemotionofthetargetistooSinalltobedetected111tl1elattercase,itispossibletodetectthetargetbya句ustingtl1eframerate

ofthecamera.九ealgo1itlunaccomplishestlrisautomaticallybyanal沪ngthe

proceedingframesuntilamajordifferenceisdetected.Wedesigt1edavoting

metl1odtoverifythetargetsbasedonaprioriknowledgeofthetargets.For小ecaseofvehicle detectionweusedvertical and horizontal gradients tolocateinteresting

featuresaswellasconstraintonareaof小etargetasdiscussedintl1si

2.1Backgroundmotionestimation

section.

Affineti·ansfo,mation[8]hasbeenusedtomodelmotionofthecamern.Thismodelincludes

rotation,scalingandti-anslation.'.?

~.Daffineti-ansfo1mationisdescribedasfollow

（