毕业论文外文翻译-基于视觉的矿井救援机器人场景识别

《毕业论文外文翻译-基于视觉的矿井救援机器人场景识别》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A英文原文ScenerecognitionforminerescuerobotlocalizationbasedonvisionCUIYi-an(崔益安),CAIZi-xing(蔡自兴),WANGLu(王璐)Abstract：AnewscenerecognitionsystemwaspresentedbasedonfuzzylogicandhiddenMarkovmodel(HMM)thatcanbeappliedinminerescuerobotlocalizationduringemergencies.Thesystemusesmonocularcameratoa

2、cquireomni-directionalimagesofthemineenvironmentwheretherobotlocates.Byadoptingcenter-surrounddifferencemethod,thesalientlocalimageregionsareextractedfromtheimagesasnaturallandmarks.TheselandmarksareorganizedbyusingHMMtorepresentthescenewheretherobotis,andfuzzylogicstrategyisusedtomatchth

3、esceneandlandmark.Bythisway,thelocalizationproblem,whichisthescenerecognitionprobleminthesystem,canbeconvertedintotheevaluationproblemofHMM.Thecontributionsoftheseskillsmakethesystemhavetheabilitytodealwithchangesinscale,2Drotationandviewpoint.Theresultsofexperimentsalsoprovethatthesystem

4、hashigherratioofrecognitionandlocalizationinbothstaticanddynamicmineenvironments.Keywords:robotlocation;scenerecognition;salientimage;matchingstrategy;fuzzylogic;hiddenMarkovmodel1IntroductionSearchandrescueindisasterareainthedomainofrobotisaburgeoningandchallengingsubject[1].Minerescuero

5、botwasdevelopedtoenterminesduringemergenciestolocatepossibleescaperoutesforthosetrappedinsideanddeterminewhetheritissafeforhumantoenterornot.Localizationisafundamentalprobleminthisfield.Localizationmethodsbasedoncameracanbemainlyclassifiedintogeometric,topologicalorhybridones[2].Withitsfe

6、asibilityandeffectiveness,scenerecognitionbecomesoneoftheimportanttechnologiesoftopologicallocalization.Currentlymostscenerecognitionmethodsarebasedonglobalimagefeaturesandhavetwodistinctstages:trainingofflineandmatchingonline.Duringthetrainingstage,robotcollectstheimagesoftheenvironmentw

7、hereitworksandprocessestheimagestoextractglobalfeaturesthatrepresentthescene.Someapproacheswereusedtoanalyzethedata-setofimagedirectlyandsomeprimaryfeatureswerefound,suchasthePCAmethod[3].However,thePCAmethodisnoteffectiveindistinguishingtheclassesoffeatures.Another