搜索
21120162JOUＲNALOFHAＲBINUNIVEＲSITYOFSCIENCEANDTECHNOLOGYVol.21No.1Feb．2016150080．．．Yoon90SSD12.34%．DOI10．15938/j．jhust．2016．01．017TP391.4A1007－2683201601－0083－06AnImprovedAdaptiveSupport-weightStereoMatchingAlgorithmwithSparseＲegionLUDiLINXueSchoolofElectricalandElectricEngineeringHarbinUniversityofScienceandTechnologyHarbin150080ChinaAbstractStereomatchingistoobtaindisparitymapfromtwoormoreimageswhosepixelsshowdisparitybe-tweeneachother．Becauseadaptivesupport-weightalgorithmisdifficulttomeetbothaccuracyandspeedanim-provedadaptivesupport-weightsparseregionapproachbasedonHVShumanvisualsystemisproposed．Firstweimprovethetraditionalsupport-weightformulaandthesupport-weightofthepointscanbecalculatedaccordingtotheimprovedsupport-weightformula．Secondthedensedisparitymapisobtainedbyusingsparseregion-basedmatching．Thirdleft-rightconsistencycheckandblockingfillingisperformedfortheobtaineddisparitymap．Final-lymedianfilterisusedtoremoveisolatedmismatchingpointsandnoisepoints．Experimentresultsshowthatbyu-singthepresentedmethodthematchingefficiencyisabove90timesfasterthanthatoftheadaptivesupport-weightalgorithmproposedbyYoonandthematchingaccuracyis12.34%higherthanthatofSSDsotheimprovedalgo-rithmisverifiedforobtainingaccuratedisparitymapatafastpaceandmeetingtherequirementsofsystempracticabil-ity．Keywordsstereomatchingproposedadaptivesupport-weightsparseregion-basedmatchingleft-rightcon-sistencycheck2015－03－101971—E-mailludizeng@hrbust．edu．cn1991—．01－8．9．10．11．12－13．14．15．．Fusiello169．Yoon17．．Yoon．3123．11.1Yoon17Yoon．．wpq=exp－Δcpqγc+Δgpqγ()()p1pqΔcpqpqΔgpqpqγcγp．Cp珋p=∑q∈Np珋q∈N珋pwpqw珋p珋qeq珋q∑q∈Np珋q∈N珋pwpqw珋p珋q2p珋pNppqNpN珋p珋p珋qN珋peq珋qq珋q．YoonSADsumofabsolutedifferences．Yoon．Yoon1．1Yoon10．．．1.2．．18．2．4821219．．．WWx+iy+j=ae－kf23Wxyf=i2+j槡2a、kakW．．W3．3W30．13W．2TADtruncatedabsolutedifferences4TADxyd=min|ILx+iy+j－IＲx+i+dy+j|T4ILIＲdTADxydT．TADabsolutedifferences．．．．5×54．44‘o’‘×’4．1.2SSWTAD=∑ij∈wLWx+iy+j×TADxyd5wL5WTAwinnertakeallSSWTAD．3581．left-rightconsistencycheckLＲC．．．41→→DL2→→DＲ3PDLPP－DLP|DLP－DＲP－DLP|＜δ4PLPＲDP=minDPLDPＲ54．5．55Middleburytsukuba．CPUAMDA6－36702.70GHz2GMatlabＲ2010b．tsukubaSSDsumofsquareddifferences17Yoonsumofweighted－supporttruncatedab-solutedifferencesSWTAD．SSD15×1515×15SSD．11×11T=6a=0.3k=1/172δ=1．．6．66atsukuba6bMiddleburytsukuba．6c15×15SSDSSD6821．6d6f6c．6d6fYoon．．PBMpercentageofbadmatchingPBMB=1N∑xy|dcxy－dTxy|＞δd．6dcxydTxyNδd1．4Middlebury．1．1MiddleburyTsukuban－occ/%All/%Disc/%/%/sYoon1.381.856.903.381719.884.945.8513.88.2018.18SWTAD5.226.1515.38.8930.71SSD10.111.941.721.2318.38n－occalldisc1n-occalldisc5．16e6fSWTAD．．．MiddleburySSD21.23%8.20%．1SSDSSD13.03%SSD．Yoon1719.88s29min．．Middlebury7．76781．LＲC12LＲC3YoonSSD4．1．J．20142251379－1387．2．J．201446566－72．3．J．20094525191－194．4．J．2012148－52．5．J．201234216－20．6STENTOUMISCGＲAMMATIKOPOULOSLKALISPEＲAKISIetal．ALocalAdaptiveApproachforDenseStereoMatchinginArchitecturalSceneＲeconstructionC//5thInternationalWork-shopon3DVirtualＲeconstructionandVisualizationofComplexArchitectures．NetherlandsElsevier2013．7．AUVJ．2009131143－148．8．J．2013184359－367．9．J．2004103－7．10．J．201116112060－2066．11．J．201116224－28．12SCHAＲSTEINDSZELISKIＲZABIHＲ．ATaxonomyandE-valuationofDenseTwo-frameStereoCorrespondenceAlgorithmsJ．InternationalJournalofComputervision20024717－42．13．J．201131344－49．14HUMENBEＲGEＲMZINNEＲCWEBEＲMetal．AFastSter-eoMatchingAlgorithmSuitableforEmbeddedＲeal-timeSystemJ．ComputerVisionandImageUnderstanding20101180－1202．15．J．2012323231－235．16FUSIELLOAＲOBEＲTOVTＲUCCOE．EfficientStereowithMultipleWindowingC//IEEEConferenceonComputerVisionandPatternＲecognition．USAIEEE1997858－863．17YOONKJKWEONIS．AdaptiveSupport-weightApproachforCorrespondenceSearchJ．IEEETrans．PatternAnalysisandMachineIntelligence2006284650－656．18NUＲGDESILVAVAＲACHCHIHKetal．SensitivityoftheHVSforBinocularDisparityCuein3DDisplaysunderDifferentAmbientIlluminationConditionsC//IEEEInternationalConfer-enceonConsumerElectronicsICCE．USAIEEE2012459－460．19KANADETOKUTOMIFM．AStereoMatchingAlgorithmwithanAdaptiveWindowTheoryandExperimentJ．IEEETrans．PatternAnalysisandMachineIntelligence1994169920－932．8821