搜索
外文翻译原文及译文学院计算机学院专业计算机科学与技术班级学号姓名指导教师负责教师沈阳航空航天大学2013年6月沈阳航空航天大学毕业设计（外文翻译）1Projectionand3DVisionInthischapterwe’llmoveintothree-dimensionalvision,frstwithprojectionsandthenwithmulticamerastereodepthperception.Todothis,we’llhavetocarryalongsomeoftheconceptsfromChapter11.We’llneedthecamerainstrinsicsmatrixM,thedistortioncoeffcients,therotationmatrixR,thetranslationvectorT,andespeciallythehomogra-phymatrixH.We’llstartbydiscussingprojectionintothe3Dworldusingacalibratedcameraandreviewingatneandprojectivetransforms(whichwefirstencounteredinChapter6);thenwe’llmoveontoanexampleofhowtogetabird’s-eyeviewofagroundplane.We’llalsodiscussPOSIT,analgorithmthatallowsustofindthe3Dpose(positionandrotation)ofaknown3Dobjectinanimage.Wewillthenmoveintothethree-dimensionalgeometryofmultipleimages.Ingeneral,thereisnoreliablewaytodocalibrationortoextract3Dinformationwithoutmultipleimages.Temostobviouscaseinwhichweusemultipleimagestoreconstructathree-dimensionalsceneisstereovision.Instereovision,teaturesintwo(ormore)imagestakenatthesametimefromseparatecamerasarematchedwiththecorrespondingtea-turesintheotherimages,andthedifferencesareanalyzedtoyielddepthinformation.Anothercaseisstructurefrommotion.Inthiscasewemayhaveonlyasinglecamera,butwehavemultipleimagestakenatdifferenttimesandfromdifferentplaces.Intheformercaseweareprimarilyinterestedindisparityeffects(triangulation)asameansofcomputingdistance.Inthelatter,wecomputesomethingcalledthefundamentalmatrix(relatestwodifferentviewstogether)asthesourceofoursceneunderstanding.Let’sgetstartedwithprojection.Projectionand3DVisionOncewehavecalibratedthecamera(seeChapter11),itispossibletounambiguouslyprojectpointsinthephysicalworldtopointsintheimage.Tismeansthat,givenalocationinthethree-dimensionalphysicalcoordinateframeattachedtothecamera,wecancomputewhereontheimager,inpixelcoordinates,anexternal3Dpointshouldap-pear.ThistransformationisaccomplishedbytheOpenCVroutinecvProjectPoints2().voidcvProjectPoints2(constCvMat*object_points,constCvMat*rotation_vector,constCvMat*translation_vector,constCvMat*intrinsic_matrix,constCvMat*distortion_coeffs,CvMat*image_points,CvMat*dpdrot=NULL,CvMat*dpdt=NULL,CvMat*dpdf=NULL,CvMat*dpdc=NULL,CvMat*dpddist=NULL,doubleaspectRatio=0);沈阳航空航天大学毕业设计（外文翻译）2Atfirstglancethenumberofargumentsmightbealittleintimidating,butinfactthisisasimplefunctiontouse.ThecvProjectPoints2()routinewasdesignedtoaccommodatethe(verycommon)circumstancewherethepointsyouwanttoprojectarelocatedonsomerigidbody.Inthiscase,itisnaturaltorepresentthepointsnotasjustalistofloca-tionsinthecameracoordinatesystembutratherasalistoflocationsintheobject’sownbodycenteredcoordinatesystem;thenwecanaddarotationandatranslationtospecifytherelationshipbetweentheobjectcoordinatesandthecamera’scoordinatesystem.Infact,cvProjectPoints2()isusedinternallyincvCalibrateCamera2(),andofcoursethisisthewaycvCalibrateCamera2()organizesitsowninternaloperation.AlloftheoptionalargumentsareprimarilythereforusebycvCalibrateCamera2(),butsophisticatedusersmightfindthemhandyfortheirownpurposesaswell.Thefirstargument,object_points,isthelistofpointsyouwantprojected;itisjustanN-by-3matrixcontainingthepointlocations.Youcangivetheseintheobject’sownlocalcoordinatesystemandthenprovidethe3-by-1matricesrotation_vectorandtranslation_vectortorelatethetwocoordinates.Ifinyourparticularcontextitiseasiertoworkdirectlyinthecameracoordinates,thenyoucanjustgiveobject_pointsinthatsystemandsetbothrotation_vectorandtranslation_vectortocontain0s.Theintrinsic_matrixanddistortion_coeffsarejustthecameraintrinsicinformationandthedistortioncoefficientsthatcomefromcvCalibrateCamera2()discussedinChap-ter11.Theimage_pointsargumentisanN-by-2matrixintowhichtheresultsofthecomputationwillbewritten.Finally,thelonglistofoptionalargumentsdpdrot,dpdt,dpdf,dpdc,anddpddistareallJacobianmatricesofpartialderivatives.Thesematricesrelatetheimagepointstoeachofthedifferentinputparameters.Inparticular:dpdrotisanN-by-3matrixofpartialde-rivativesofimagepointswithrespecttocomponentsoftherotationvector;dpdtisanN-by-3matrixofpartialderivativesofimagepointswithrespecttocomponentsofthetranslationvector;dpdfisanN-by-2matrixofpartialderivativesofimagepointswithrespecttofxandfy;dpdcisanN-by-2matrixofpartialderivativesofimagepointswithrespecttocxandcy;anddpddistisanN-by-4matrixofpartialderivativesofimagepointswithrespecttothedistortioncoefficients.Inmostcases,youwilljustleavetheseasNULL,inwhichcasetheywillnotbecomputed.Telastparameter,aspectRatio,isalsooptional;itisusedforderivativesonlywhentheaspectratioisfxedincvCalibrateCamera2()orcvStereoCalibrate().Ifthisparameterisnot0thenthederivativesdpdfareadjusted.AffineandPerspectiveTransformationsTwotransformationsthatcomeupoftenintheOpenCVroutineswehavediscussed—aswellasinotherapplicationsyoumightwriteyourself—aretheaffineandperspectivetransformations.WefirstencounteredtheseinChapter6.AsimplementedinOpenCV,theseroutinesaffecteitherlistsofpointsorentireimages,andtheymappointsononelocationintheimagetoadifferentlocation,offenperformingsubpixelinte