Robust wide baseline stereo from maximally stable extremal r(4)
发布时间:2021-06-07
发布时间:2021-06-07
The wide baseline stereo problem ,i.e.the problem of establishing correspondences between apair of images taken from different view points is studied.A new set of image elements that are put into correspondence,the so called extremal regions,is introduced.Extremal regions possess highly desirable properties
764J.Matasetal./ImageandVisionComputing22(2004)761–767
Fig.3.Wash.Epipolargeometryanddensematchedregionswithfullyaf ne
distortion.
stablethresholdsexistandasystemofnestedsubsetsisoutputinthiscase.FinallyweremarkthatMSERscanbede nedonanyimage(evenhigh-dimensional)whosepixelvaluesarefromatotallyorderedset.
3.Theproposedrobustwide-baselinealgorithmDistinguishedregiondetection.Asa rststep,theDRsaredetected—theMSERscomputedontheintensityimage(MSERþ)andontheinvertedimage(MSER2).
Fig.4.EstimatedEGonanoutdoor
scene.
Measurementregions.Ameasurementregionofarbi-trarysizemaybeassociatedwitheachDR,iftheconstructionisaf ne-covariant.Smallermeasurementregionsarebothmorelikelytosatisfytheplanarityconditionandnottocrossadiscontinuityindepthororientation.Ontheotherhand,smallregionsarelessdiscriminative,i.e.theyaremuchlesslikelytobeunique.Increasingthesizeofameasurementregioncarriestheriskofincludingpartsofbackgroundthatarecompletelydifferentinthetwoimagesconsidered.Clearly,theoptimalsizeofaMRdependsonthescenecontentanditisdifferentforeachDR.In[21],TuytelaarsandVanGooldoubletheellipticalDRtoincreasediscriminability,whilekeepingtheprobabilityofcrossingobjectboundariesatanacceptablelevel.
Intheproposedalgorithm,measurementregionsareselectedatmultiplescales:theDRitself,1.5,2and3timesscaledconvexhulloftheDR.Sincematchingisaccomplishedinarobustmanner,webene tfromtheincreaseofdistinctivenessoflargeregionswithoutbeingseverelyaffectedbyclutterornon-planarityoftheDR’monly,MahalanobisdistancehasbeenusedinMRmatching.However,thenon-robustnessofthismetricmeansthatmatchingmayfailbecauseofasinglecorruptedmeasure-ment(thishappenedintheexperimentsreportedbelow).Invariantdescription.Inallexperiments,rotationalinvariants(basedoncomplexmoments)[8]wereusedafterapplyingatransformationthatdiagonalisestheregionscovariancematrixoftheDR.Incombination,thisisanaf binationofrotationalandaf nelyinvariantgeneralisedcolourmoments[12]gaveasimilarresult.Ontheirown,theaf neinvariantsfailedonproblemswithalargescalechange.
Robustmatching.Ameasurementtakenfromanalmostplanarpatchofthescenewithstableinvariantdescriptionwillbereferredtoasa‘goodmeasurement’.Unstablemeasurementsorthosecomputedonnon-planarsurfaces
or
Fig.5.Cylindricalbox.Epipolargeometry(top)andmatchedregions(bottomleft).Fullyaf nedistortion,anon-planarobject,texturedsurfaceandstrongspecularre ectionsarepresentinthescene.SHOUT(bottomright),ascenewithachangeofilluminationspectralpowerdistribution.
上一篇:设计素描教案
下一篇:基于模块化的自动装配方法