[ros-users] tabletop_object_detector, table and object not segmented

Mon Jan 10 17:20:52 UTC 2011

Hi Ugo,

If I use this instead of "fixed":
> <param name="processing_frame" value="openni_camera" />
>
> then it works properly.
>

Well, glad it's fixed; I can venture some guesses on what was wrong below
and let us know if you start seeing problems again.

It is true that the "base_link" frame is considered a "default frame" in
parts of the code. There is need for one, as well as a frame in which
assumptions can be made regarding which way is "logical down" (gravity),
etc. We have tried to make that a parameter, but we might have missed it in
various places where it's hard coded in. I would not be very surprised if we
had strange behaviors in a system with no "base link" frame. However, that
is indeed the right place to specify your own processing frame, so I'm
encouraged that changing that parameter fixed your problem.

> > As you can see on those screenshots, though the table plan is found
> > correctly, the segmented point cloud includes the table and the object.
> >
> > http://img5.imagebanana.com/img/xx8ltmur/Selection_001.png
> > http://img5.imagebanana.com/img/ji89gls7/Selection_002.png
>

I would guess this is related to the problem above. The point clouds are
coming in in a frame that is not the processing frame, and "base_link" was
not available to transform into. I agree that it should however give you
some more informative error messages instead of wrong point clouds...

> >
> > I'm not sure what I did wrong. In the log below, I see those warnings:
> > [ INFO] [1294654799.697276843]: cannot extend bbox to table, they are
> > not in the same frame
> > [ WARN] [1294654799.697396094]: Failed to extend bbox to table; using
> > original dimensions
>

This is a different part of the code, in the
tabletop_collision_map_processing. In there, we take the bounding boxes of
segmented point clouds and add them to the collision system. We have found
that even if the point cloud appears to not extend all the way down to the
table, the system is more robust if we force its bounding box to extend to
the table (since it is unlikely that the object is floating in mid-air).
That process fails due to the same problem above, but the error correction
catches it and just gives you the original bounding box, not extended to
touch the table. Should not be a stopper for anything downstream.

Best,
Matei

> > My only clue is that lots of code in tabletop_object_detector and the
> > depending packages seems to have "base_link" hard coded as the reference
> > link. I'm using "fixed", so I'm publishing a fixed transform between
> > fixed and base_link. (I tried to use: <remap from="base_link"
> > to="fixed" /> in my nodes but I get the error: [ERROR] [WallTime:
> > 1294655504.963270] tf transform was not there!!). Anyway I don't think
> > that it's related to my problem, as even with this error, I get the
> > yellow rectangle for the table + the object segmented point cloud which
> > contains the table as shown previously.
> >
> > Sorry if my message is not precise enough, but I'm a bit lost here and
> > I'm not sure where to start looking.
> >
> > Any help greatly appreciated.
> >
> > Cheers,
> >
> > Ugo
> >
> >
> > [ INFO] [1294654796.985860596]: Tabletop detection service called;
> > waiting for a point_cloud2 on topic /camera/depth/points2
> > [ INFO] [1294654797.261886506]: Point cloud received; processing
> > [ INFO] [1294654797.374996037]: Input cloud converted to fixed frame
> > [ INFO] [1294654797.375096220]: Starting process on new cloud
> > [ INFO] [1294654797.400421841]: Step 1 done
> > [ INFO] [1294654797.433487674]: Step 2 done
> > [ INFO] [1294654797.464712409]: [TableObjectDetector::input_callback]
> > Model found with 2992 inliers: [0.090311 -0.057030 0.994279 -0.009768].
> > [ INFO] [1294654797.464846818]: Step 3 done
> > [ INFO] [1294654797.465131387]: Step 4 done
> > [ INFO] [1294654797.477030009]: Table computed
> > [ INFO] [1294654797.508113609]:  Number of object point candidates:
> > 55342.
> > [ INFO] [1294654798.549648017]: Number of clusters found matching the
> > given constraints: 1.
> > cloud_objects frame: fixed
> > [ INFO] [1294654798.568788369]: Clusters converted
> > [ INFO] [1294654798.595098187]: Fitting models to clusters
> > [ INFO] [1294654799.025706263]:   - fit with score 0.046118 for model id
> > 18791
> > [ INFO] [1294654799.063682509]: Cloud size: 34522
> > [INFO] [WallTime: 1294654799.444427] finding the bounding box for a
> > point cluster
> > [INFO] [WallTime: 1294654799.613361] waiting for transform: base_link
> > from fixed
> > [INFO] [WallTime: 1294654799.617360] got transform
> > [INFO] [WallTime: 1294654799.645909] chopped points off of dim -0,
> > lowest val = -0.353, searchval = -0.327
> > [INFO] [WallTime: 1294654799.668446] chopped points off of dim -1,
> > lowest val = -0.298, searchval = -0.274
> > [INFO] [WallTime: 1294654799.690813] chopped points off of dim 2,
> > highest val = 0.308, searchval = 0.261
> > [ INFO] [1294654799.697276843]: cannot extend bbox to table, they are
> > not in the same frame
> > [ WARN] [1294654799.697396094]: Failed to extend bbox to table; using
> > original dimensions
> > [ INFO] [1294654799.697444127]: Adding bounding box with dimensions
> > 0.623375 0.423401 0.260162 to collision map
> > [ERROR] [1294654829.700053341]: Collision map was not formed in allowed
> > time
> > [ERROR] [1294654829.700812477]: Collision map processing error;
> > exception: collision map: static make was not formed in allowed time
> >
>
> --
> Ugo Cupcic | Shadow Robot Company | ugo at shadowrobot.com
> Software Engineer | 251 Liverpool Road |
> need a Hand? | London N1 1LX | +44 20 7700 2487
> http://www.shadowrobot.com/hand/ @shadowrobot
>
> _______________________________________________
> ros-users mailing list
> ros-users at code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users
>

-- 
Matei Ciocarlie
Research Scientist
Willow Garage Inc.
650-475-9780
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ros.org/pipermail/ros-users/attachments/20110110/cbd6dc11/attachment-0003.html>