Hi,
we're using the camera_pose_calibration package and try to calibrate multiple camera's with it. We followed the instructions at http://www.ros.org/wiki/camera_pose_calibration - which went well. The camera nodes are started, the images appear properly in the calibration gui, but we have trouble detecting the checkerboard pattern. Although the pattern appears, clearly visible and sharp, in both images, no corners are detected (see screenshot attached C:\fakepath\Screenshot-13.png). We tried this with 2 kinect camera's at first, and with 2 prosilica Gige camera's afterwards (we thought image quality might matter here, but it's no different). We tested with both mono and color images. Any suggestions / comments on how to check what is going wrong here?
Hi Steven,
Did you ensure to specify the checker_rows and checker_cols parameters in your top-level launchfile? Note that the checkerboard that ships with the PR2 has 7x6 corners (not 8x7).
What resolution are you running the cameras at? If your images are significantly greater than 640x480, then the checkerboard detectors are going to start to choke pretty badly. You can see how fast each checkerboard detector is running with:
> rostopic hz [camera_ns]/features
In our setup, we're using binning to downsample the images, and hacked camera_info to make the images look like binning was off. It might make sense for us to make the binning functionality fully supported by our tools.
Steven,
The development version (hg clone https://kforge.ros.org/calibration/camera_pose) provides better user feedback about what's going on in the calibration pipeline. This should help you find out what's going wrong.
Asked: 2011-05-03 02:05:14 -0500
Seen: 665 times
Last updated: May 03 '11
ROS Answers is licensed under Creative Commons Attribution 3.0 Content on this site is licensed under a Creative Commons Attribution Share Alike 3.0 license.
