From pcl-users-bounces@code.ros.org Tue Jul 26 14:38:23 2011 Return-Path: X-Original-To: www-data@code.ros.org Delivered-To: www-data@code.ros.org Received: from pub5.willowgarage.com (localhost [127.0.0.1]) by pub5.willowgarage.com (Postfix) with ESMTP id 5BD4B680C6; Tue, 26 Jul 2011 14:38:23 -0700 (PDT) X-Original-To: pcl-users@code.ros.org Delivered-To: pcl-users@code.ros.org Received: from mail-gy0-f174.google.com (mail-gy0-f174.google.com [209.85.160.174]) by pub5.willowgarage.com (Postfix) with ESMTPS id 47A0525C851 for ; Tue, 26 Jul 2011 14:38:22 -0700 (PDT) Received: by gyh3 with SMTP id 3so787794gyh.5 for ; Tue, 26 Jul 2011 14:38:21 -0700 (PDT) MIME-Version: 1.0 Received: by 10.142.249.30 with SMTP id w30mr3919448wfh.122.1311716300716; Tue, 26 Jul 2011 14:38:20 -0700 (PDT) Received: by 10.143.66.2 with HTTP; Tue, 26 Jul 2011 14:38:20 -0700 (PDT) In-Reply-To: <4E2F2ABD.9010507@willowgarage.com> References: <4E26FC04.3060509@willowgarage.com> <4E274CC5.9020306@willowgarage.com> <4E2F262D.3000808@willowgarage.com> <4E2F2ABD.9010507@willowgarage.com> Date: Tue, 26 Jul 2011 14:38:20 -0700 Message-ID: From: Suat Gedikli To: "Radu B. Rusu" , "Point Cloud Library (PCL) mailing list" Subject: Re: [PCL-users] Segment planes perpendicular to axis Z X-BeenThere: pcl-users@code.ros.org X-Mailman-Version: 2.1.9 Precedence: list Reply-To: "Point Cloud Library \(PCL\) mailing list" List-Id: "Point Cloud Library \(PCL\) mailing list" List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Content-Type: multipart/mixed; boundary="===============6941896306109514497==" Mime-version: 1.0 Sender: pcl-users-bounces@code.ros.org Errors-To: pcl-users-bounces@code.ros.org --===============6941896306109514497== Content-Type: multipart/alternative; boundary=001636ed6ac5267b9f04a8ffc088 --001636ed6ac5267b9f04a8ffc088 Content-Type: text/plain; charset=ISO-8859-1 Just a short question: do you want to use RANSAC for that? since thats not required if you know that the planes are perpendicular to the z-axis. All points with similar z-values lie on a plane => you need just a clustering in the 1-dimensional array consisting of the z-values of all points. The plane parameters are then (0, 0, 1, -Zavg), with Zavg being the average of all z-values within the cluster. By programming this explicitly its should be much faster than any SAC method, and also deterministic (depending on the clustering method used). Or did i miss the point? -Suat On Tue, Jul 26, 2011 at 1:59 PM, Radu B. Rusu wrote: > Aaron, > > Sure. If I'm not mistaken, r1893 and r1894 are relevant: > > http://dev.pointclouds.org/projects/pcl/repository/revisions/1893 > http://dev.pointclouds.org/projects/pcl/repository/revisions/1894 > > Cheers, > Radu. > -- > Point Cloud Library (PCL) - http://pointclouds.org > > On 07/26/2011 01:58 PM, Aaron O'Toole wrote: > > Radu, > > > > Would you mind letting me know which files you changed in response to > this? > > > > Thanks, > > > > > > On Tue, Jul 26, 2011 at 4:40 PM, Radu B. Rusu rusu@willowgarage.com>> wrote: > > > > Ruben, > > > > I believe I fixed this in trunk and all branches. Can you please give > it a try again? > > > > Thank you in advance. > > > > Cheers, > > Radu. > > -- > > Point Cloud Library (PCL) - http://pointclouds.org > > > > On 07/22/2011 07:01 AM, Ruben Ares wrote: > > > I've tried with a point cloud with 100 points, all of them with > z=1. > > > > > > The next code (with eps angle corrected) doesn't find any plane. > > > > > > g.setModelType (pcl::SACMODEL_PERPENDICULAR_PLANE); > > > seg.setMethodType (pcl::SAC_RANSAC); > > > seg.setMaxIterations(1000); > > > seg.setDistanceThreshold (0.1); > > > seg.setAxis(Eigen::Vector3f(0,0,1)); > > > seg.setEpsAngle (0.1); > > > > > > I'm going to create an issue on the tracker. > > > > > > Thanks for the help. > > > > Date: Wed, 20 Jul 2011 14:46:45 -0700 > > > > From: rusu@willowgarage.com > > > > To: pcl-users@code.ros.org > > > > CC: bladecsp@hotmail.com > > > > Subject: Re: [PCL-users] Segment planes perpendicular to axis Z > > > > > > > > Ruben, > > > > > > > > > > > > > > > > On 07/20/2011 02:26 PM, Ruben Ares wrote: > > > > > Are there any restrictions in the segmentation module about > the coordinate system? I'm only need planes > > > perpendicular to > > > > > my Z-axis and I don't understad why the coordinate system must > be relevant. > > > > > > > > Nope, there are no restrictions. Ok, so you want planes > perpendicular to your camera's Z. That's easy. > > > > > > > > > The scene has that kind of planes but the next code doesn't > find any of them. > > > > > > > > > > seg.setModelType (pcl::SACMODEL_PERPENDICULAR_PLANE); > > > > > seg.setMethodType (pcl::SAC_RANSAC); > > > > > seg.setMaxIterations(1000); > > > > > seg.setDistanceThreshold (0.05); > > > > > seg.setAxis(Eigen::Vector3f(0,0,1)); > > > > > seg.setEpsAngle (0); > > > > > > > > I think the last call might be the reason for that. You specify > an absolute threshold of "0" radians, which > > means you > > > > want the plane to be perfect. We seem to have introduced a > condition that says that the angle has to be larger > > than 0 > > > > here (check sac_model_perpendicular_plane.hpp, line 85 in > trunk), which could be both a bug and a feature :) > > (meaning > > > > that we're flexible with respect to changing this condition). > > > > > > > > As far as I remember the reason for this check is that we need a > bit of wiggle room for the sample consensus > > methods in > > > > terms of getting their samples, otherwise we risk at spending > more time in the loop (overconstraining the > > system). This > > > > should not affect the end results, meaning that if your plane is > perfectly perpendicular you will still get it. > > The only > > > > time this assumption would break is if you would have 2 planes > within that tolerance threshold with the exact > > number of > > > > points, and for some reason you want to differentiate between > them. > > > > > > > > So basically you need so specify a small tolerance (0.01 radians > is good enough if you know for a fact that > > your plane's > > > > equation is perfectly perpendicular to Z) in order for this to > kick in. I just updated the documentation to > > point this > > > > thing. Sorry for not mentioning this earlier. > > > > > > > > If you still have problems with this, can you please create an > issue on our tracker > > (http://dev.pointclouds.org) with a > > > > simple dataset and a minimal code example so we can reproduce > the error and try to fix it? Thanks. > > > > > > > > Cheers, > > > > Radu. > > > > > > > > > _______________________________________________ > > > PCL-users@code.ros.org / > http://pointclouds.org > > > https://code.ros.org/mailman/listinfo/pcl-users > > _______________________________________________ > > PCL-users@code.ros.org / > http://pointclouds.org > > https://code.ros.org/mailman/listinfo/pcl-users > > > > > > > > > > -- > > - Aaron O'Toole > _______________________________________________ > PCL-users@code.ros.org / http://pointclouds.org > https://code.ros.org/mailman/listinfo/pcl-users > --001636ed6ac5267b9f04a8ffc088 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable Just a short question: do you want to use RANSAC for that? since thats not = required if you know that the planes are perpendicular to the z-axis. All p= oints with similar z-values lie on a plane =3D> you need just a clusteri= ng in the 1-dimensional array consisting of the z-values of all points. The= plane parameters are then (0, 0, 1, -Zavg), with Zavg being the average of= all z-values within the cluster.
By programming this=A0explicitly=A0its should be much faster than any SAC m= ethod, and also deterministic (depending on the clustering method used).
Or did i miss the point?

-Suat

On Tue, Jul 26, 2011 at 1:59 PM, Radu B= . Rusu <rusu@= willowgarage.com> wrote:
Aaron,

Sure. If I'm not mistaken, r1893 and r1894 are relevant:

http://dev.pointclouds.org/projects/pcl/repository/rev= isions/1893
http://dev.pointclouds.org/projects/pcl/repository/rev= isions/1894

Cheers,
Radu.
--
Point Cloud Library (PCL) - http://pointclouds.org

On 07/26/2011 01:58 PM, Aaron O'Toole <airuno2L@gmail.com> wrote:
> Radu,
>
> Would you mind letting me know which files you changed in response to = this?
>
> Thanks,
>
>
> On Tue, Jul 26, 2011 at 4:40 PM, Radu B. Rusu = <rusu@willowgarage.com <= mailto:rusu@willowgarage.com&g= t;> wrote:
>
> =A0 =A0 Ruben,
>
> =A0 =A0 I believe I fixed this in trunk and all branches. Can you plea= se give it a try again?
>
> =A0 =A0 Thank you in advance.
>
> =A0 =A0 Cheers,
> =A0 =A0 Radu.
> =A0 =A0 --
> =A0 =A0 Point Cloud Library (PCL) - http://pointclouds.org
>
> =A0 =A0 On 07/22/2011 07:01 AM, Ruben Ares wrote:
> =A0 =A0 =A0> I've tried with a point cloud with 100 points, all= of them with z=3D1.
> =A0 =A0 =A0>
> =A0 =A0 =A0> The next code (with eps angle corrected) doesn't f= ind any plane.
> =A0 =A0 =A0>
> =A0 =A0 =A0> g.setModelType (pcl::SACMODEL_PERPENDICULAR_PLANE); > =A0 =A0 =A0> seg.setMethodType (pcl::SAC_RANSAC);
> =A0 =A0 =A0> seg.setMaxIterations(1000);
> =A0 =A0 =A0> seg.setDistanceThreshold (0.1);
> =A0 =A0 =A0> seg.setAxis(Eigen::Vector3f(0,0,1));
> =A0 =A0 =A0> seg.setEpsAngle (0.1);
> =A0 =A0 =A0>
> =A0 =A0 =A0> I'm going to create an issue on the tracker.
> =A0 =A0 =A0>
> =A0 =A0 =A0> Thanks for the help.
> =A0 =A0 =A0> > Date: Wed, 20 Jul 2011 14:46:45 -0700
> =A0 =A0 =A0> > From: rusu@willowgarage.com <mailto:rusu@willowgarage.com>
> =A0 =A0 =A0> > To: pcl= -users@code.ros.org <mailto:pcl-users@code.ros.org>
> =A0 =A0 =A0> > CC: blade= csp@hotmail.com <mailto:blad= ecsp@hotmail.com>
> =A0 =A0 =A0> > Subject: Re: [P= CL-users] Segment planes perpendicular to axis Z
> =A0 =A0 =A0> >
> =A0 =A0 =A0> > Ruben,
> =A0 =A0 =A0> >
> =A0 =A0 =A0> >
> =A0 =A0 =A0> >
> =A0 =A0 =A0> > On 07/20/2011 02:26 PM, Ruben Ares wrote:
> =A0 =A0 =A0> > > Are there any restrictions in the segmentati= on module about the coordinate system? I'm only need planes
> =A0 =A0 =A0> perpendicular to
> =A0 =A0 =A0> > > my Z-axis and I don't understad why the = coordinate system must be relevant.
> =A0 =A0 =A0> >
> =A0 =A0 =A0> > Nope, there are no restrictions. Ok, so you want = planes perpendicular to your camera's Z. That's easy.
> =A0 =A0 =A0> >
> =A0 =A0 =A0> > > The scene has that kind of planes but the ne= xt code doesn't find any of them.
> =A0 =A0 =A0> > >
> =A0 =A0 =A0> > > seg.setModelType (pcl::SACMODEL_PERPENDICULA= R_PLANE);
> =A0 =A0 =A0> > > seg.setMethodType (pcl::SAC_RANSAC);
> =A0 =A0 =A0> > > seg.setMaxIterations(1000);
> =A0 =A0 =A0> > > seg.setDistanceThreshold (0.05);
> =A0 =A0 =A0> > > seg.setAxis(Eigen::Vector3f(0,0,1));
> =A0 =A0 =A0> > > seg.setEpsAngle (0);
> =A0 =A0 =A0> >
> =A0 =A0 =A0> > I think the last call might be the reason for tha= t. You specify an absolute threshold of "0" radians, which
> =A0 =A0 means you
> =A0 =A0 =A0> > want the plane to be perfect. We seem to have int= roduced a condition that says that the angle has to be larger
> =A0 =A0 than 0
> =A0 =A0 =A0> > here (check sac_model_perpendicular_plane.hpp, li= ne 85 in trunk), which could be both a bug and a feature :)
> =A0 =A0 (meaning
> =A0 =A0 =A0> > that we're flexible with respect to changing = this condition).
> =A0 =A0 =A0> >
> =A0 =A0 =A0> > As far as I remember the reason for this check is= that we need a bit of wiggle room for the sample consensus
> =A0 =A0 methods in
> =A0 =A0 =A0> > terms of getting their samples, otherwise we risk= at spending more time in the loop (overconstraining the
> =A0 =A0 system). This
> =A0 =A0 =A0> > should not affect the end results, meaning that i= f your plane is perfectly perpendicular you will still get it.
> =A0 =A0 The only
> =A0 =A0 =A0> > time this assumption would break is if you would = have 2 planes within that tolerance threshold with the exact
> =A0 =A0 number of
> =A0 =A0 =A0> > points, and for some reason you want to different= iate between them.
> =A0 =A0 =A0> >
> =A0 =A0 =A0> > So basically you need so specify a small toleranc= e (0.01 radians is good enough if you know for a fact that
> =A0 =A0 your plane's
> =A0 =A0 =A0> > equation is perfectly perpendicular to Z) in orde= r for this to kick in. I just updated the documentation to
> =A0 =A0 point this
> =A0 =A0 =A0> > thing. Sorry for not mentioning this earlier.
> =A0 =A0 =A0> >
> =A0 =A0 =A0> > If you still have problems with this, can you ple= ase create an issue on our tracker
> =A0 =A0 (http= ://dev.pointclouds.org) with a
> =A0 =A0 =A0> > simple dataset and a minimal code example so we c= an reproduce the error and try to fix it? Thanks.
> =A0 =A0 =A0> >
> =A0 =A0 =A0> > Cheers,
> =A0 =A0 =A0> > Radu.
> =A0 =A0 =A0>
> =A0 =A0 =A0>
> =A0 =A0 =A0> _______________________________________________
> =A0 =A0 =A0> = PCL-users@code.ros.org <mailto:PCL-users@code.ros.org> / http://pointclouds.org
> =A0 =A0 =A0> https://code.ros.org/mailman/list= info/pcl-users
> =A0 =A0 _______________________________________________
> =A0 =A0 PCL-users@code= .ros.org <mailto:PCL-users= @code.ros.org> / http://pointclouds.org
> =A0 =A0 https://code.ros.org/mailman/listinfo/pcl= -users
>
>
>
>
> --
> - Aaron O'Toole

--001636ed6ac5267b9f04a8ffc088-- --===============6941896306109514497== Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Disposition: inline _______________________________________________ PCL-users@code.ros.org / http://pointclouds.org https://code.ros.org/mailman/listinfo/pcl-users --===============6941896306109514497==--