Re: [Ros-kinect] Point cloud registration with Kinect

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

Re: [Ros-kinect] Point cloud registration with Kinect

Radu B. Rusu
Administrator
Christian,

We're working on revamping the Registration framework to work efficiently with the Kinect as we speak. Please join us on
pcl-users@ for discussions.

We should have a demo and something working properly by the end of the week.

Cheers,
Radu.
--
http://pointclouds.org

On 12/13/2010 11:19 AM, cmuell2s wrote:

> Hi,
>
> I have a question about point_cloud registration with kinect.
>
> Did somebody tried to apply ICP (pcl::IterativeClosestPointnonLinear<pcl::PointYXZ,pcl::PointXYZ>) from the pcl library
> on kinect 3D point clouds?
>
> I tried to register several poind_clouds from a table-top area to get a better table top representation.
> The kinect is tilted with 2 degress per shot.
>
> The current point_cloud is aligned with the last registrated point_cloud (pairwise registration approach).The aligned
> current point_cloud is added to a final_point_cloud which is supposed to represent the model of the table top scene.
>
> --Before applying ICP : statisticalOutlierRemoval<pcl::PointXYZ> is applied to the current point_cloud (meanK=50,
> stddevMulThresh=1.0f).
> --The maximum iterations of ICP is set to 100, and the transformationEpsilon is set to 1e-8.
> --Also I apply down sampling to the current point_cloud (Leafsize is 0.02f from the VoxelGrid filter)
>
> The alignment is quite bad as you can see in the attached screen shot (see yellow circle - it is supposed to be a single
> plane, namely the table).
> Also I added some small code sample of the point_cloud registration part.
>
> Do somebody has experiences with kinect+icp? I wonder whether the resulted point_cloud is too unstable, by means there
> are to many distortions in the point_clouds from the same tilt-angle. Or did I have overlooked something?
>
> Thanks for any comment!
>
> Cheers,
>
> Christian
>
>
>
>
>
> _______________________________________________
> Ros-kinect mailing list
> [hidden email]
> https://code.ros.org/mailman/listinfo/ros-kinect
_______________________________________________
[hidden email] / http://pointclouds.org
https://code.ros.org/mailman/listinfo/pcl-users
Reply | Threaded
Open this post in threaded view
|

Re: [Ros-kinect] Point cloud registration with Kinect

garratt
We've also been working on this, as well as using vslam to get better alignment through bundling.  How have you been attacking this problem?

Garratt

On Mon, Dec 13, 2010 at 2:22 PM, Radu Bogdan Rusu <[hidden email]> wrote:
Christian,

We're working on revamping the Registration framework to work efficiently with the Kinect as we speak. Please join us on
pcl-users@ for discussions.

We should have a demo and something working properly by the end of the week.

Cheers,
Radu.
--
http://pointclouds.org

On 12/13/2010 11:19 AM, cmuell2s wrote:
> Hi,
>
> I have a question about point_cloud registration with kinect.
>
> Did somebody tried to apply ICP (pcl::IterativeClosestPointnonLinear<pcl::PointYXZ,pcl::PointXYZ>) from the pcl library
> on kinect 3D point clouds?
>
> I tried to register several poind_clouds from a table-top area to get a better table top representation.
> The kinect is tilted with 2 degress per shot.
>
> The current point_cloud is aligned with the last registrated point_cloud (pairwise registration approach).The aligned
> current point_cloud is added to a final_point_cloud which is supposed to represent the model of the table top scene.
>
> --Before applying ICP : statisticalOutlierRemoval<pcl::PointXYZ> is applied to the current point_cloud (meanK=50,
> stddevMulThresh=1.0f).
> --The maximum iterations of ICP is set to 100, and the transformationEpsilon is set to 1e-8.
> --Also I apply down sampling to the current point_cloud (Leafsize is 0.02f from the VoxelGrid filter)
>
> The alignment is quite bad as you can see in the attached screen shot (see yellow circle - it is supposed to be a single
> plane, namely the table).
> Also I added some small code sample of the point_cloud registration part.
>
> Do somebody has experiences with kinect+icp? I wonder whether the resulted point_cloud is too unstable, by means there
> are to many distortions in the point_clouds from the same tilt-angle. Or did I have overlooked something?
>
> Thanks for any comment!
>
> Cheers,
>
> Christian
>
>
>
>
>
> _______________________________________________
> Ros-kinect mailing list
> [hidden email]
> https://code.ros.org/mailman/listinfo/ros-kinect
_______________________________________________
[hidden email] / http://pointclouds.org
https://code.ros.org/mailman/listinfo/pcl-users


_______________________________________________
[hidden email] / http://pointclouds.org
https://code.ros.org/mailman/listinfo/pcl-users
Reply | Threaded
Open this post in threaded view
|

Re: [Ros-kinect] Point cloud registration with Kinect

Radu B. Rusu
Administrator
PS. We haven't forgotten about this -- we've just been busy preparing the NI Kinect/PrimeSense stack for its first
release. Once it's out, we'll continue the discussion. :)

Cheers,
Radu.

On 12/13/2010 11:48 AM, garratt gallagher wrote:
> We've also been working on this, as well as using vslam to get better alignment through bundling.  How have you been
> attacking this problem?
>
> Garratt
_______________________________________________
[hidden email] / http://pointclouds.org
https://code.ros.org/mailman/listinfo/pcl-users