Ask Your Question

Pointclouds extraction using the Bumblebee2 stereo camera

asked 2013-04-04 09:42:57 -0600

INSA gravatar image

updated 2014-01-28 17:16:03 -0600

ngrennan gravatar image

Hi guys,

This is my first results using the bumblebee2 and the "stereo_image_proc" node to extract pointclouds. I calibrated the camera with the "camera_calibration" node using the -k 6 option and got an epi ratio of around 0.25. link: (PointCloud)

What do you think about this results ? is it possible to have a better calibration and a better pointcloud ? what are the best calibration conditions ?
Is there any way to process this pointcloud to get the same result than this one : link: (viso2_ros)

Stephan, can you please tell me how did you process your pointcloud (The one in the video below) and what epi ratio you got after the calibration of your Bb2 ? Does your "viso2_ros" perform the Slam that we can see in this same video ? link: (viso2_Video)

Thank you guys for your responses

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted

answered 2013-04-04 10:23:18 -0600

Stephan gravatar image

updated 2013-04-04 10:28:12 -0600

Your clouds look very good!

Regarding the video on the viso2_ros wiki page: The object is at a distance of about 1m and it has a very good texture. To get the reconstruction, we downsampled the point clouds using VoxelGridFilter and accumulated them in rviz using viso2_ros to generate odometry, setting the fixed frame in rviz to /odom and the decay time of the PointCloud2 display to 10 seconds. You can use fovis_ros for the visual odometry as well. There is no SLAM algorithm in the background so the reconstruction will suffer from drift after a couple of seconds. I don't remember what we got as epipolar error during our calibration. I think 0.25 is okay.


If you want to slow down processing (stereo_image_proc can get quite busy on calculating point clouds) you can throttle one input topic for stereo_image_proc.

edit flag offensive delete link more


Thank you Stephan for your help. What about the blank areas in the pointcloud ? Specially when i move the camera. How can i fill these empty spaces ? maybe with Temporal Filtering ? Your Pointcloud has no empty spaces...

INSA gravatar image INSA  ( 2013-04-04 21:45:08 -0600 )edit

I suppose the texture is not unique enough for block matching. You could try to adapt the stereo_image_proc parameters correlation_window_size and/or uniqueness_ratio. In the end, if there is no texture the algorithm is unable to match the area from left to right image. Paint your walls! ;-)

Stephan gravatar image Stephan  ( 2013-04-05 06:32:03 -0600 )edit

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Question Tools

1 follower


Asked: 2013-04-04 09:42:57 -0600

Seen: 1,212 times

Last updated: Apr 04 '13