欢迎来到天天文库
浏览记录
ID:34594968
大小:1.99 MB
页数:8页
时间:2019-03-08
《Deep learning of spatio-temporal features with geometric-based moving point detection for motion segmentation副本.pdf》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、2014IEEEInternationalConferenceonRobotics&Automation(ICRA)HongKongConventionandExhibitionCenterMay31-June7,2014.HongKong,ChinaDeepLearningofSpatio-TemporalFeatureswithGeometric-basedMovingPointDetectionforMotionSegmentationTsung-HanLinandChieh-ChihWangAbstract—Thispaperintroducesanapproachtoacco
2、mplishSpatio-temporalFeaturesLearningmotionsegmentationfromamovingstereocamerabasedondeeplearning.Previousworkonmovingobjectdetectionmostlyusepointfeaturesbasedon3Dgeometricconstraints.However,pointfeaturesrequiregoodfeatures,andarehardtodetectortobematchedcorrectlyinsituationswhereobjectshavesm
3、oothtextures.Toalleviatethisproblem,learninghigh-levelspatio-RNNtemporalfeaturesunsupervisedlyfromrawimagedatabasedSegmentationonReconstructionIndependentComponentAnalysis(RICA)autoencodersisproposed.Despitethepowerofthenewspatio-MotionSegmentationtemporalfeatures,thesefeaturescannotnotlearnandb
4、eusedStereoImagestointerpret3Dgeometryofdynamicscenes,whichiscriticalEgomotionformovingobjectdetectionfrommovingcameras.Asdetected3DGeometric-basedmovingpointsmovingpointsbasedon3Dgeometricconstraintsstillcontainvaluableinformationof3Dsceneaswellasthecameraego-motion,weproposeaframeworkthatincor
5、poratesboththeFig.1:Theproposeddeeplearningframeworkformotiondetectedmovingpointresultsandthelearnedspatio-temporalsegmentationfromamovingstereocamera.featuresasinputstoRecursiveNeuralNetworks(RNN)thatperformsmotionsegmentation.Bothfeatureseffectivelycom-plementeachother.Theproposedapproachisdem
6、onstratedwithreal-worldstereovideodatathatcontainsmultiplemovingInthiswork,wefocusonlearningmorerobusthigh-objects,andhasachieved26%betterdetectionrateoverthelevelmotionfeaturesbasedondeeplearningtoimproveexisting3Dgeometric-basedmovingpointsdetector.movingobjectdetectionatthefront-endbeforeanyp
7、ostprocessing.OurworkusesanetworkbasedonReconstruc-I.INTRODUCTIONtionIndependentComponentAnalysis(RICA)autoencoderDetectingmovingobjectssuchascarsandpedestriansis[5][6]tounsupervisedlylearnhigh-levelspatio-temporalimportantf
此文档下载收益归作者所有