No subject

not enough sample images to compute the correct calibration.  I'd suggest
gathering more images before hitting the 'calibrate' button.  Try to get the
checkerboard:

   - filling the entire frame
   - very distant, in the four corners of the frame
   - both face-on and skewed for the above cases

Then hit 'calibrate'.  The gui uses a simple heuristic for when it thinks it
has enough calibration data, which seems to predict quite well for the PR2's
stereo cameras.  It is likely that your custom setup is sufficiently
different that the heuristic is being overly optimistic.

J

On Mon, Oct 11, 2010 at 6:57 AM, Jeff Hyams <jeff.hyams at resquared.com>wrote:

> I am trying to run the camera_calibration to calibrate a simple custom
> setup stereo pair.  I am using the uvc_stereo package to publish the raw
> camera images.
>
> I get through the calibration (using a 5x4 checkerboard at the moment)
> after a minute of moving the checkerboard to the various positions around
> the image.  But once the calibrate button lights up and I hit it, it seems
> to still be looking for the checkerboard.  It also shows images which look
> to be zoomed in samples of the actual images, rather than what I would
> expect for poorly calibrated rectified images.  It also seems to be
> constantly recalculating the epipolar error (really high, usually over 10.0)
> whenever it finds the checkerboard after I have hit the calibrate button, or
> reporting the checkerboard not found if it's not in the image.
>
> Based on the wiki page, I expected it to sample the images, then once the
> button was hit, to calculate and then print out or save the results.  Is
> this expected behavior, or am I doing something drastically wrong here?  Am
> I just getting a really bad set of images for calibration that is screwing
> up the calculations?
>
> I am only running the image_raw publisher(s).  Do I need to also be running
> the stereo_image_proc while doing calibration?
>
> Thanks,
>
> Jeff Hyams
> RE2, Inc.
>
> _______________________________________________
> ros-users mailing list
> ros-users at code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users
>
>


-- 
J.

--001636e90310b4ae25049259a8c8
Content-Type: text/html; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable