ICCV2011-initial_registration.pdf

《ICCV2011-initial_registration.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::RegistrationInitialAlignmentof3DPointCloudsDirkHolzNovember6,2011RegistrationRegistering3DPointCloudsforbuilding3Dmodelsofobjects,tablescenes,andwholerooms/buildings/outdoorenvironments.1.Initialalignment(registrationwithoutpreviousknowledge)2.Iterativerefinement(givenaninitialalignme

2、nt)3.Multi-viewregistrationDirkHolzPointCloudLibrary(PCL)RegistrationRegistering3DPointCloudsIGivenasourcepointcloudandatargetpointcloud1.determinecorrespondencepairs,2.estimateatransformationthatalignsthecorrespondences,3.applythetransformationtoalignsourceandtarget.Icorrespondencescanbe

3、anything(points/features/...)!DirkHolzPointCloudLibrary(PCL)RegistrationRegistrationPipelineinPCLDirkHolzPointCloudLibrary(PCL)InitialAlignment(1/9)1.Computesetsofkeypoints2.Compute(local)featuredescriptors(e.g.FPFH)3.Matchfeaturestofindcorrespondences4.Estimatetransformationfromcorrespond

4、encesPersistentFeaturePointsHistogramsPersistentFeaturePointsHistogramsPersistentFeaturePointsHistograms353535P1P2P330303025Q125Q225Q3202020151515101010Ratioofpointsinonebin(%)5Ratioofpointsinonebin(%)5Ratioofpointsinonebin(%)5000024681012141602468101214160246810121416BinsBinsBinsDirkHolz

5、PointCloudLibrary(PCL)InitialAlignment(2/9)Code:MatchingFeaturesCorrespondenceEstimationest;est.setInputCloud(source_features);est.setInputTarget(target_features);est.determineCorrespondences(correspondences);Example:Foundcorrespondences/matches$correspondence_viewerrob

6、ot0robot1-n50DirkHolzPointCloudLibrary(PCL)InitialAlignment(3/9)TransformationEstimationISeveralmethodsforcomputingatransformationT=(R;t)givencorrespondencepairs(di;mi):IPoint-to-pointIPoint-to-planeIPlane-to-planeI...andmanyothersISimplesolution(basedonSVD)forminimizingpoint-to-pointdist

7、ance(leastsquareserrorE):X2E(T)=(mi(Rdi+t))ipcl::registration::TransformationEstimationSVDDirkHolzPointCloudLibrary(PCL)InitialAlignment(4/9)Code:TransformationestimationEigen::Matrix4ftransformation;TransformationEstimationSVDsvd;svd.estimateRigidTr