预览加载中，请您耐心等待几秒...

基于双目视觉的无监督深度感知及语义分割.docx

预览
1/3
2/3
3/3

在线预览结束，喜欢就下载吧，查找使用更方便

下载提示 文本预览

如果您无法下载资料，请参考说明：

1、部分资料下载需要金币，请确保您的账户上有足够的金币

2、已购买过的文档，再次下载不重复扣费

3、资料包下载后请先用软件解压，在使用对应软件打开

基于双目视觉的无监督深度感知及语义分割 Abstract Depthperceptionandsemanticsegmentationaretwofundamentaltasksincomputervision.However,mostexistingmethodsfordepthperceptionandsemanticsegmentationrelyonsupervisedlearningandrequirelargeamountsoflabeleddata.Inthispaper,weproposeanovelapproachbasedonstereovisionforunsuperviseddepthperceptionandsemanticsegmentation.Weleveragethenaturalabilityofhumanvisiontoperceivedepthusingbinocularvisionandcombineitwithdeeplearningtechniquestoaddressthesetasks.Specifically,weutilizeastereocamerasetuptocapturestereoimages,andthenuseadeepneuralnetworktolearnthedepthinformationandsemanticlabelsinanunsupervisedmanner.Experimentalresultsonthebenchmarkdatasetdemonstratetheeffectivenessofourproposedapproach. 1.Introduction Depthperceptionandsemanticsegmentationareimportanttasksincomputervision,withwideapplicationsinautonomousdriving,robotics,andaugmentedreality.Depthperceptioncalculatesthedistancebetweenthecameraandobjectsinthescene,whilesemanticsegmentationassignssemanticlabelstoeachpixelintheimage.Althoughthesetasksareessentialforunderstandingthevisualworld,theyarechallengingduetothecomplexityandvariabilityofreal-worldscenes. Traditionalmethodsfordepthperceptionoftenrelyonstereomatchingorstructurefrommotionalgorithms.Thesemethodsrequirethecalibrationofcamerasandassumetheavailabilityofdepthmapsor3Dmodelsfortraining.Thislimitstheirapplicabilityinreal-worldscenarioswherelabeleddataisscarceorexpensivetoobtain.Ontheotherhand,superviseddeeplearningapproacheshaveachievedgreatsuccessindepthperceptionandsemanticsegmentation.However,theyheavilyrelyonlargeamountsoflabeledtrainingdata,whichcanbetime-consumingandcostlytoacquire. Inthispaper,weproposeanunsupervisedapproachfordepthperceptionandsemanticsegmentationbasedonstereovision.Weaimtoleveragethenaturalabilityofhumanvisiontoperceivedepthusingbotheyesandcombineitwithdeeplearningtechniquestoaddressthesetasksinanunsupervisedmanner.Thekeyideabehindourapproachistolearndepthandsemanticlabelsfromunlabeledstereoimages,withoutthe