ICCV2011-features.pdf

《ICCV2011-features.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、PCL::FeaturesMichaelDixonandSuatGedikliNovember6,2011Outline1.Introduction2.FeatureEstimation3.KeypointDetection4.CombiningKeypointsandFeaturesPointCloudLibrary(PCL)IntroductionInthissession,we’lltalkaboutfeaturesandkeypointsinPCL.ILocalfeatureestimationISurfacenormal/curvaturees

2、timationIPointFeatureHistogramdescriptorsIKeypointdetectionI3DSIFTkeypointdetectionIHarris-liekeypointdetectionIGlobalfeatureestimationIViewpointFeatureHistogramdescriptorsPointCloudLibrary(PCL)Outline1.Introduction2.FeatureEstimation3.KeypointDetection4.CombiningKeypointsandFeat

3、uresPointCloudLibrary(PCL)LocalFeaturesWhatisafeature?Whatisafeature?Invision/perception,theword“feature”canmeanmanydifferentthings.InPCL,“featureestimation”means:IcomputingafeaturevectorbasedoneachpointslocalneighborhoodIorsometimes,computingasinglefeaturevectorforthewholecloudP

4、ointCloudLibrary(PCL)LocalFeaturesWhatisafeature?Featurevectorscanbeanythingfromsimplesurfacenormalstothecomplexfeaturedescriptorsneedforregistrationorobjectdetection.Today,we’lllookatacoupleofexamples:ISurfacenormalestimation(NormalEstimation)IPointFeatureHistogramestimation(FPF

5、HEstimation)IViewpointFeatureHistogramestimation(VFHEstimation)PointCloudLibrary(PCL)LocalFeatures(1-2/5)SurfaceNormalEstimation.TheoreticalAspects:BasicIngredientsIGivenapointcloudwithx,y,z3DpointcoordinatesPointCloudLibrary(PCL)LocalFeatures(1-2/5)SurfaceNormalEstimation.Theore

6、ticalAspects:BasicIngredientsIGivenapointcloudwithx,y,z3DpointcoordinatesISelecteachpoint’sk-nearestneighbors,fitalocalplane,andcomputetheplanenormalPointCloudLibrary(PCL)LocalFeatures(3/5)SurfaceNormalandCurvatureEstimationPkT1PkCj=i=1i
(pipj)
(pipj);p=k
i=1pi82

7、pioutlier0+1+2i:1;piinlierPointCloudLibrary(PCL)FeatureestimationbasicsICreatethefeatureestimationobjectI(Templatedontheinputandoutputpointtypes)ISetthesearchmethodtousefortheunderlyingneighborhoodsearchesICreateak-DTreeandpassittothefeatureestimatorISpecifythesizeofthelocalne

8、ighborhoodtousewhencomputingthefeaturesI