资源描述:
《基于DSP的通过局部特征实时物体识别嵌入式系统 毕业论文外文翻译》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、附录AReal-timeobjectrecognitionusinglocalfeaturesonaDSP-basedembeddedsystemAbstractInthelastfewyears,objectrecognitionhasbecomeoneofthemostpopulartasksincomputervision.Inparticular,thiswasdrivenbythedevelopmentofnewpowerfulalgorithmsforlocalappearancebasedobj
2、ectrecognition.So-called‘‘smartcameras’’withenoughpowerfordecentralizedimageprocessingbecamemoreandmorepopularforallkindsoftasks,especiallyinthefieldofsurveillance.Recognitionisaveryimportanttoolastherobustrecognitionofsuspiciousvehicles,personsorobjectsisa
3、matterofpublicsafety.Thissimplymakesthedeploymentofrecognitioncapabilitiesonembeddedplatformsnecessary.Inourworkweinvestigatethetaskofobjectrecognitionbasedonstate-of-the-artalgorithmsinthecontextofaDSP-basedembeddedsystem.Weimplementseveralpowerfulalgorith
4、msforobjectrecognition,namelyaninterestpointdetectortogetherwithanregiondescriptor,andbuildamedium-sizedobjectdatabasebasedonavocabularytree,whichissuitableforourdedicatedhardwaresetup.Wecarefullyinvestigatetheparametersofthealgorithmwithrespecttotheperform
5、anceontheembeddedplatform.Weshowthatstate-of-the-artobjectrecognitionalgorithmscanbesuccessfullydeployedonnowadayssmartcameras,evenwithstrictlylimitedcomputationalandmemoryresources.KeywordsDSP;Objectrecognition;Localfeatures;Vocabularytree1.IntroductionObj
6、ectrecognitionisoneofthemostpopulartasksinthefieldofcomputervision.Inthepastdecade,bigeffortsweremadetobuildrobustobjectrecognitionsystemsbasedonappearancefeatureswithlocalextent.Forsuchaframeworktobeapplicableintherealworldseveralattributesareveryimportant
7、:insensitivityagainstrotation,illuminationorviewpointchanges,aswellasreal-timebehaviorandlarge-scaleoperation.Currentsystemsalreadyhavealotofthesepropertiesand,thoughnotallproblemshavebeensolvedyet,nowadaystheybecomemoreandmoreattractivetotheindustryforincl
8、usioninproductsforthecustomermarket.Inturn,recentlyembeddedvisionplatformssuchassmartcamerashavesuccessfullyemerged,however,onlyofferingalimitedamountofcomputationalandmemoryresources.Nevertheless,embe