翻译原文.docx

上传人:b****6 文档编号:12069164 上传时间:2023-06-04 格式:DOCX 页数:19 大小:753.92KB
下载 相关 举报
翻译原文.docx_第1页
第1页 / 共19页
翻译原文.docx_第2页
第2页 / 共19页
翻译原文.docx_第3页
第3页 / 共19页
翻译原文.docx_第4页
第4页 / 共19页
翻译原文.docx_第5页
第5页 / 共19页
翻译原文.docx_第6页
第6页 / 共19页
翻译原文.docx_第7页
第7页 / 共19页
翻译原文.docx_第8页
第8页 / 共19页
翻译原文.docx_第9页
第9页 / 共19页
翻译原文.docx_第10页
第10页 / 共19页
翻译原文.docx_第11页
第11页 / 共19页
翻译原文.docx_第12页
第12页 / 共19页
翻译原文.docx_第13页
第13页 / 共19页
翻译原文.docx_第14页
第14页 / 共19页
翻译原文.docx_第15页
第15页 / 共19页
翻译原文.docx_第16页
第16页 / 共19页
翻译原文.docx_第17页
第17页 / 共19页
翻译原文.docx_第18页
第18页 / 共19页
翻译原文.docx_第19页
第19页 / 共19页
亲,该文档总共19页,全部预览完了,如果喜欢就下载吧!
下载资源
资源描述

翻译原文.docx

《翻译原文.docx》由会员分享,可在线阅读,更多相关《翻译原文.docx(19页珍藏版)》请在冰点文库上搜索。

翻译原文.docx

翻译原文

7.3MultiviewIntegration

Thetechniquesdescribedinthisbookforrecoveringdepthfrommatchedstereo-pairsofim-agesresultinasetofpoints,therangemap,beingacquiredthatcanbetriangulatedtodescribeasingle2.5Dmanifoldin3-space.Inmanypracticalapplicationsof3Dimaging,thereisarequirementtogenerateacompleteandclosed3Dsurfacemanifoldbyfusingtogether,i.e.integrating,multiple2.5Drangemapscorrespondingtomultipleviewscapturedofasingleobject.Inthiscasesufficientviewsoftheobjectmustbecapturedtoensurethataclosed3Dmeshcanbeformed.

CurlessandLevoy[82]stipulatethat,ideally,theaboveprocessshouldmakeuseofalltherangedatacollected,takeintoaccountthequalityorcertaintyofeachlocalrangemeasure-ment,generatethesame3Dmeshirrespectiveoftheorderinwhichrangemapsareprocessedandallowincrementaladditionofrangemaps.Inaddition,theintegrationprocessshouldalsoundertakestepstoamelioratedefectsinthecaptureddata,forexamplebydetectingandre-movingrangemapoutliersorfillingholesintheconstructed3Dmesh(totherebyconstructawatertightmesh).Finally,theprocessshouldnotberestrictedtoobjectsofanyspecific

 

topologicalconfigurationandshouldalsobecomputationallyefficient,astensorevenhun-dredsofrangemapsmighthavetobeintegratedinrealapplications.

Averylargebodyofworkonrangesurfaceintegrationhasbeenreportedintheliterature;however,twotechniquesaremostcommonlyadopted,volumetricintegrationanddirectmeshintegration,asdescribedinthefollowingsections.

 

7.3.1ImplicitSurfacesandMarchingCubes

Thebasicideabehindvolumetricintegration,asdescribedbyCurlessandLevoy[82],involvesdecimatingthe3Dspaceinwhichthecapturedrangesurfacesliebymeansofavoxeldatastructure.RepresentationandintegrationofmultiplerangesurfaceswithinindividualvoxelelementsrelyontheconceptofasigneddistancefunctionD(x)thatrecordsthedistanceinspacefromanimagedsurfacetothecentreofeachvoxel.Thisdistanceisrecordedfromeachvoxelcentretotherangesurface,followingthepathofthesightlinefromtheimagingsensor,forpositivedistances.Negativedistancesoccurwherethesensorsightlinepenetratestherangesurfacepriortoreachingaspecificvoxel.Figure7.2illustratesthisgeometricconfiguration.

Accordingly,thezero-surface,i.e.whenD(x)=0,representseachrangesurfaceand

D(x)isthereforetermedanimplicitfunction.Foreachrangesurfacetobeintegratedthecorrespondingsigneddistancefunctions,di(x),areconstructedandaccumulatedinvoxelspacetoprovideanewzero-surfacethatineffectaveragestheirrelativedisplacements.This

 

Figure7.2Signeddistancerangesurfaceencodingofvoxelspace.Figurebasedon[82]

 

accumulatedzero-surfacecanthenbeextractedandtriangulatedusingatechniquesuchasmarchingcubes[280],describedbrieflyinsection7.3.1.4.

Afurthercomplicationisthatthequalityofthephysicalsensingmechanismisalsotakenintoaccountwhenformingtheimplicitfunctionforeachrangesurface.Whenviewingasurfacefromthereferencecamera(say)ofastereo-pair,assumingfrontoparallelgeometryforsimplicity,thentheoptimumsurfacereconstructionwilltakeplacewhentheobservedsurfacenormaliscollineartothiscamera’ssightline.Acosineangleweightingfunctionwi(x)canthereforebeformedbytakingthedotproductofthe(dominant)camerasightlineunitvectorandtheobservedsurfacenormal.Thisfunctionisusedtoweighteachimplicitfunctionsampled.Itisthenpossibletosumthesampledimplicitfunctionswithineachvoxel(x)fromeachrangeimapbytakingthefirst-ordermomentsasfollows:

i

},wi(x)di(x)

D(x)=

},w(x).(7.22)

Furthermore,theabovesumcanbeformedincrementallyasfollowstogivethesigneddis-tancefunctionDi(x)andweightfunctionWi(x)accumulatedforthei-thrangemap:

Wi(x)Di(x)+wi+1(x)di+1(x)

Di+1(x)=

Wi(x)

+wi+1

(x)

(7.23)

Wi+1(x)=Wi(x)+wi+1(x).

Thedistancexinvoxelspaceoverwhichsigneddistancesareformedinfrontofandbehindtherangemaphastoberestrictedtoavoidthesurfacesofself-occludingmanifoldsfrominter-feringwitheachother.Thisplacesalimitontheminimumthicknessofclosedmanifoldthatcanbeconstructedbythismethod.IntheimplementationofCurlessandLevoytheimplicitfunctionisformedforhalfthemaximumuncertaintyintervalintherangemeasurementsinfrontof(andbehind)therangemapsurface.

 

7.3.1.1RangeMapPre-segmentation

Duetothecharacteristicsofrangemapscapturedbystereo-photogrammetryanumberofpreprocessingstepsareusuallyrequiredtoassistsegmentingacleanlyclosedvolumefrommultiplecapturedrangemapsofanobject.Anotherreasonforsegmentingoutonlysurfacesofinterestisduetothen3memorycostofconstructingavoxelspace,wherenisthesamplesizeofeachdimensionoftheimagedvolume,itisvitaltoreducetheimagedvolumesizetoencompassonlyvaliddata.

Itisnotuncommontoadopt‘bluescreen’coloursegmentationtoisolatetheobjectsurfaceofinterestintheintensityimagesofthestereo-pairsandthenusetheseassegmentationmaskstoisolatethecorrespondingrelevantareaintherangemapsforintegration.Clearlythisimposesthelimitationofcapturingthedesiredobjectinfrontofanappropriatelycolouredbackdrop,typicallyblue,greenororange,thatdoesnotcorrespondtothecolourofthesurfacetobereconstructed.Asthissegmentationprocessmayresultinfragmentationoftherangemap,usuallyanumberofmorphologicaloperationsareappliedtosmooththeresultantbinarysegmentationmask,selectthelargestcontiguousblob(assumedtocorrespondtothesurfaceofinterest)andthenfillanyholesremaininginthisblob.

 

Whencoloursegmentationisnotpossibleorinappropriate(monochromestereo-pairsmightonlybeavailableoritmightnotbepossibletoconstrainthecaptureconditions)thenitispossibletothresholdtheconfidencemapproducedbythestereomatchingalgorithminordertoidentifycontiguousregionscomprisingviablymatchedsurfaces.

7.3.1.2VolumetricIntegrationAlgorithmOverview

TheprincipalbasicalgorithmicstepssetoutbyCurlessandLevoy[82]areasfollows.

1.Initializethevoxelspacewithzeros.

2.Constructtrianglesonthenearestneighbourelementsoftherangemapssuchthattrianglesarenotformedoversteepdiscontinuitiesbydetectingtriangleswhosesidelengths,whentakenasratiosofeachother,exceedathresholdlimit.Inthismannerthesurfacenormalforeachtrianglecanbeextractedanditsdotproductformedwiththe(dominant)cameralineofsighttoproduceaweightvalueforeachobservedelementofeachrangesurface.

3.Thesigneddistanceforeachvoxelforeachrangemapiscomputedbycastingarayfrom

theprincipalpointinthedominantcamerathrougheachelementintherangemap,andthedistancenotedtoeachvoxelwithinadistanceof±Dmaxvoxelsasdeterminedbytherange

measurementuncertainty.

4.EachvoxelelementisupdatedbyaccumulatingtheweightedsigneddistancesusingEquation(7.24).

5.IsolateanewisosurfaceforD(x)=0.

7.3.1.3HoleFilling

Whiletheabovealgorithmwillgenerateagoodapproximationtoanisosurface,unseenareascanresultinholeswhenattemptingtotriangulatetheisosurface.Sincethereareoftensitua-tionswherethepresenceofholesinasurfaceisunacceptable,e.g.whenmeasuringvolumechangeduetosurfacedisplacement,acommonclinicalrequirementwhenassessingtheout-comeofcertaintherapiesorsurgeryproceduresonthebodyorface,ameansoffillingholesisrequiredtoproduceawatertightsurfacereconstruction.

Thestandardapproachistolabelthevoxelsaccordingtooneofthefollowingstates:

unseen,emptyorsurface(within±Dmaxoftheaccumulatedsurface).Figure7.3illustrates

thisapproach,anditcanbeseenthatsurfaceholesariseattheboundaryofemptyandunseenvoxels.Therefore,placingsurfacesattheseboundariesprovidesasimplemeansofgeneratingawatertightsurfacebasedontheminimumofassumptions.

Thepreviouslydescribedalgorithmisnowmodifiedasfollows.

1.Assignallvoxelstoaninitialunseenstate.

2.Computetheweightedsigneddistanceaccumulationonthosevoxelswithin±Dmaxofthe

accumulatedsurfacesasbefore,labellingthesevoxelsassurface.

3.Resetallthevoxelsbetweenthoselabelledassurface,followingthedominantcameraviewingdirection,backtotheboundaryofthevoxelspaceclosesttothiscamerawiththestateempty.

4.OncemoreextracttheisosurfaceforD(x)=0andalsoextractasurfaceattheinterface

betweenthoseregionslabelledunseenandthoseregionslabelledempty.

 

Unseen

HoleFillIsosurface

 

SignedDistance±d

Empty

 

ObservedIsosurface

 

Sensor

 

Figure7.3Geometricconfigurationforholefilling.Figurebasedon[82]

 

Whereholefillingisappliedthensurfacediscontinuitiesarelikelytoarise,therebygen-eratingartefacts.Thisobviouseffectofsuchholefillingcanbereducedbyapplyinglocalfiltering(averagingnearestneighbouringvertices)onlytotheholeregionsthemselves.Inthiswayitispossibletoavoidblurringtheremainderofthemeshwhilesuppressingartefacts.Thesupportregionofthefilterweightsisallowedtotaperbetweenthehole-filledregionandtheobserveddatainordertosmooththeirtransition[82].

 

7.3.1.4MarchingCubes

Thefinalstepoftheabovevolumetricintegrationmethodrequiresthattheisosurfacefor

D(x)=0isconstructedfromthevoxelscontainingtheac

展开阅读全文
相关资源
猜你喜欢
相关搜索
资源标签

当前位置:首页 > 高等教育 > 医学

copyright@ 2008-2023 冰点文库 网站版权所有

经营许可证编号:鄂ICP备19020893号-2