文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > 工程科技 > 电子/电路 > Dense Correspondence Finding for Parametrization-free Animation
侵权投诉

Dense Correspondence Finding for Parametrization-free Animation

DenseCorrespondenceFindingforParametrization-freeAnimation

ReconstructionfromVideo

NaveedAhmedMPIInformatik

ChristianTheobaltStanfordUniversity

ChristianR¨osslMagdeburgUniversity

SebastianThrunStanfordUniversity

Hans-PeterSeidelMPIInformatik

Abstract

Wepresentadense3Dcorrespondence ndingmethod

thatenablesspatio-temporallycoherentreconstructionofsurfaceanimationsfrommulti-viewvideodata.Givenasin-putasequenceofshape-from-silhouettevolumesofamov-ingsubjectthatwerereconstructedforeachtimeframein-dividually,ourmethodestablishesdensesurfacecorrespon-dencesbetweensubsequentshapesindependentlyofsurfacediscretization.Thisisachievedintwosteps: rst,weob-tainsparsecorrespondencesfromrobustopticalfeaturesbetweenadjacentframes.Second,wegeneratedensecor-respondenceswhichserveasmapbetweenrespectivesur-faces.Byapplyingthisproceduresubsequentlytoallpairsoftimestepswecantriviallyalignoneshapewithallothers.Thus,theoriginalinputcanbereconstructedasasequenceofmesheswithconstantconnectivityandsmalltangentialdistortion.Weexemplifytheperformanceandaccuracyofourmethodusingseveralsyntheticandcapturedreal-worldsequences.

1.Introduction

Inrecentyears,evermoreef cientcomputersandin-creasinglyaccurateimagingdeviceshaverendereditfeasi-bletocapturecomputeranimationsfromsubjectsperform-inginthereal-worldratherthanbyhand-craftingthemwiththetraditionaltoolboxoftheanimator.Tothisend,ava-rietyofmethodshavebeendevelopedthatreconstructbothtime-varyingshapeandappearanceofarbitraryreal-worldperformersfrommulti-viewpointvideo,Sect.2.

Mostofthesemethodsprovideconvincingshapeandappearanceforeachtimestepofaninputanimationindi-vidually.However,theyfallshortofreconstructingspatio-temporallycoherentscenegeometryforarbitrarysubjectssincethechallenging3Dcorrespondenceproblemisnotaddressed.Spatio-temporalcoherenceisanimportantandhighly-desirablepropertyincapturedanimations,asitgreatlyfacilitatesorevenisinevitableformanytaskssuchasediting,compressionorspatio-temporalpostprocessing.

978-1-4244-2243-2/08/$25.00 ©2008 IEEE

Wethereforeproposeanewspatio-temporaldense3Dcorrespondence ndingmethodthatenablesustocapturecoherentdynamicscenegeometryusingstandardshape-from-silhouettemethods,Sect.3.Ouralgorithmistailoredtothecharacteristicsofvideo-basedreconstructionmethodswhichoftencapturehighspatialdetailintheinputvideoframes,butproviderelativelysparselysampled3Dgeome-trywithamuchlowerlevelofshapedetailandwithacon-siderablelevelofnoise.

Ina rststep,shape-from-silhouettesurfacesarerecon-structedforeachtimestepofvideoyieldingasequenceofshapesmadeoftrianglemesheswithvaryingconnectivity.Thereafter,sparse3Dcorrespondencesbetweensubsequentpairsofsurfacesarecomputedbymatching3Dpositionsofopticalfeaturesthatcanbeaccuratelyextractedfromhigh-resolutioninputvideoframes,Sect.3.1.Thesesparsecorre-spondencesrepresentcontrolpointsforanchoringappropri-atebivariatescalarfunctionsoneachreconstructedsurfacemesh,Sect.3.2.Thechoiceofthesefunctionsenablesustoestablishdensecorrespondenceessentiallybymatchingfunctionvalues.Thedensecorrespondencescanbeusedtostraightforwardlyalignonemeshtoallotherreconstruc-tionsbyperformingasequenceofpairwiseregistrations,Sect.3.3.Theoutputofourapproachisaspatio-temporallycoherentanimation,i.e.asequenceofmesheswithconstantgraphstructureandlowtangentialdistortion.

2.RelatedWork

Technologicalprogressinrecentyearshasmadeitfeasibletoreconstructshapeandappearanceofdynamicscenesusingvideo[16]orvideoplusactivesensing[28].Multi-viewvideomethodsbasedontheshape-from-silhouette[17]orstereoprinciple[30]beartheintriguingadvantagethattheyenablereconstructionofarbitrarymov-ingsubjects.Unfortunately,Noneofthesemethodsisde-signedtoreconstructscenegeometrywithcoherentcon-nectivityovertimesincethe3Dcorrespondenceproblemisnotaddressed.Model-basedapproachesemployshapepriors[7,6]whichlimitsthemtocertaintypesofscenes.Thealgorithmproposedinthispaperenablescoherentdy-

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email