预览加载中，请您耐心等待几秒...

基于机器学习的点集匹配算法（英文）.docx

预览
1/2
2/2

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

基于机器学习的点集匹配算法（英文） Introduction Pointsetmatchingisafundamentalproblemincomputervisionandrobotics,whichaimstoaligntwoormoresetsofpoints.Solvingthisproblemhasmanyapplications,suchas3Dreconstruction,objectrecognition,andposeestimation.Inrecentyears,machinelearninghasshowngreatpotentialinpointsetmatching,whereitcanlearntheunderlyingstructureofthedataandthematchingfunction.Inthispaper,wewillreviewseveralpopularmachinelearning-basedpointsetmatchingalgorithmsandtheirapplications. Methodology Thepointsetmatchingproblemcanbedividedintotwocategories:rigidandnon-rigidmatching.Rigidmatchingreferstoaligningtwopointsetswithoutchangingtheirgeometry,whereasnon-rigidmatchingallowsfordeformationofoneorbothpointsets.Machinelearning-basedalgorithmsforpointsetmatchingcanbeclassifiedintosupervisedandunsupervisedmethods. Supervisedmethodsrequirealabeleddataset,wherethegroundtruthcorrespondencesaregiven.Themainideaistotrainamodeltopredictthecorrespondencesbetweenthetwopointsets.Theinputfeaturescanberawgeometricinformation,suchaspointcoordinatesandnormalvectors,orotherhigh-levelfeatures,suchasshapedescriptors.Theoutputcanbeabinaryclassificationlabelindicatingwhethertwopointsmatchornot,oracontinuousvaluerepresentingthesimilarityscorebetweenthetwopoints.Examplesofsupervisedmethodsincludedeeplearning-basedalgorithms,suchasPointNetandPointNet++. Unsupervisedmethods,ontheotherhand,donotrequirelabeleddataandrelyonthegeometricpropertiesofthepointsets.Themainideaistolearnalatentspacerepresentationofthepointsets,whereacertaindistancemetriccanbecalculatedtomeasurethesimilaritybetweentwopointsets.Examplesofunsupervisedmethodsincludetheiterativeclosestpoint(ICP)algorithm,wherepointsetsareiterativelyalignedbyminimizingthedistancebetweentheclosestpoints,andspectralgraphmatching,whichconstructsagraphrepresentationofthepointsetsandfindsthecorrespondencesbetweenthembasedonthegraphspectralproperties. Applications Themachinelearning-basedpointsetmatchingalgorithmshaveshowngreatsuccessinvariousapplications.Oneo