Machine_Learning

《Machine_Learning》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、IROS2012VilaMoura,Algarve,PortugalPCL::MachineLearning–TreesandFernsStefanHolzer,TUMunich(TUM)October13,2012OverviewGoalfortoday:MachineLearninginPCL-IntroductiontoDecisionTreesandFerns-Howto:LearnanduseDecisionTrees-Howto:LearnanduseFernsMotivationComputingfeaturesfromorganizeddataisco

2、mputationallyexpensive!estimatingcurvatureRGBimagedetectingkeypointsdepthmapfindingobjectpartsetc…matchingkeypointsetc…Trees/Ferns:•Basedonsimpleandfastcomparisons•Canbeeasilyparallized(e.g.onGPUs)PCL-MLPCL-ML:PCL´sMachineLearningLibrary–DecisionTrees•Asksadifferentquestionateachnode–Fe

3、rnsClassifier•Asksabunchofquestions.atthesametime.....PCL-ML:DecisionTreesToy-Example:CurvatureEstimationWhatdoweneed?•TrainingData•Tree-LearningProcedure•Tree-EvaluationProcedurePCL-ML:DecisionTreesCreatingTrainingData:RecordPointCloudsComputeCurvaturesTrainDecisionTreeusingKinectPCL-M

4、L:DecisionTreesBasicStructureofaDecisionTree:root-node•Treestartswitharoot-nodePCL-ML:DecisionTreesBasicStructureofaDecisionTree:??•Treestartswitharoot-node•Decisionsbasedondepthcomparisonsxx12x1israndompositionx2israndompositionNxNwindowPCL-ML:DecisionTreesBasicStructureofaDecisionTree

5、:??•Treestartswitharoot-node•Decisionsbasedondepthcomparisons•Thresholdisappliedondepthdifferencetomakedecisionxx12x1israndompositionx2israndompositionNxNwindowPCL-ML:DecisionTreesBasicStructureofaDecisionTree:??•Treestartswitharoot-node•Decisionsbasedondepthcomparisons•Thresholdisappli

6、edondepthdifferencetomakedecision•x,xandthresholdareselectedduringtraining12xx12x1israndompositionx2israndompositionNxNwindowPCL-ML:DecisionTreesBasicStructureofaDecisionTree:•Treestartswitharoot-node•Decisionsbasedondepthcomparisons•Thresholdisappliedondepthdifferencetomakedecision•x,x

7、andthresholdareselectedduringtraining12•Decisionsaretrackeddowntillleafleafxx12x1israndompositionx2israndompositionNxNwindowPCL-ML:DecisionTreesBasicStructureofaDecisionTree:•Treestartswitharoot-node•Decisionsbasedondepthcomparisons•Thresholdisappliedondepthdifferencetomakedeci