[ros-users] Navigation stack: issue about ground plane detection

Radu Bogdan Rusu rusu at willowgarage.com
Wed Oct 27 07:14:11 UTC 2010


I might be able to help with that until Eitan wakes up (like he ever sleeps anyway).

Can you send me the dataset as a BAG/PCD file (1 PointCloud only please) and tell me what (and how) nodes are you using 
to filter the data precisely?


On 10/27/2010 12:09 PM, clarkwu wrote:
> Eiten,
> I did some further test about the ground plane removal on our robot.
> With the zero position of tilt laser calibrated, now I am able to see that
> before ground removal, cloud points corresponding to the ground are located
> roughly on the horizontal plane, except a few discrete points still left on
> the ground plane after filtering.
> Another strange issue is that if the "planar_refine" value set to "1" in
> launch file, the ground removal filter will not only remove the ground, but
> also a large part of points whose z value may be as high as 0.4 meter. I
> traced back into the source file sac_inc_ground_removal_standalone.cpp and
> found that it's due to the  plane fitting module within subroutine
> "computePointNormal". The calculation result from this module about the
> fitted plane seems incorrect. But I can not tell exactly which step inside
> is wrong since I extracted the code within it into a separate file and
> calculate some sample points, and the result is matching with the that from
> matlab.
> Please see enclosed screenshots for details and your comments are welcome.
> best regards
> xiaojun
> screeshot without planar refinement, the bottom edge of clouds is close to
> the ground plane:
> http://ros-users.122217.n3.nabble.com/file/n1777999/Screenshot_noplanarrefine_0.png
> screeshot without planar refinement, a few discrete points stay on the
> ground after removal
> http://ros-users.122217.n3.nabble.com/file/n1777999/Screenshot_noplanarrefine_1.png
> screenshot with planar refinement, a large part of points on top of ground
> have been removed:
> http://ros-users.122217.n3.nabble.com/file/n1777999/Screenshot_planarrefine_1.png

More information about the ros-users mailing list