ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question

visual odometry (rpg_svo vs viso2)

asked 2016-05-10 17:50:17 -0500

jys gravatar image

Does anyone have an experience between these two visual odometry packages? (e.g. for a wide-angle camera attached to an end of robot arm) The paper also does not seem talk about it.

edit retag flag offensive close merge delete



if you have chosen viso2_ros for visual odometry, could you offer me which guides/tutorials/readmes you used to set it up?

Behsad Ghanei gravatar image Behsad Ghanei  ( 2016-05-25 06:04:58 -0500 )edit

1 Answer

Sort by ยป oldest newest most voted

answered 2018-01-27 12:34:00 -0500

ZanaZak gravatar image


Before comparison, I must mention that if you want to use one single camera, your localization would be up to a scale. Even the method that monocular viso2 uses for scale estimation holds for ground vehicles and tries to estimate ground plane. I don't think an arm robot convey such a condition!

BUT regardless of scale,

rpg_svo is much more newer and efficient than viso2. It also benefits from local bundle adjustment which makes it more accurate. The only limitation is that rpg_svo was designed for downward looking cameras. Fortunately, they have recently released SVO 2.0 that handles forward looking as well as stereo and multi-camera systems. But bear in mind that SVO is a direct method for visual odometry. All direct methods, like LSD SLAM and DSO, need global shutter cameras. Rolling shutters induce huge geometric distortions which can deteriorate accuracy. If you use a rolling shutter camera, increase the frame rate to reduce this effect.

edit flag offensive delete link more


Thanks for the insight -- very helpful. Do you have a recommendation for a stereo odometry package that tracks features?

clyde gravatar image clyde  ( 2018-03-10 17:30:09 -0500 )edit

Question Tools



Asked: 2016-05-10 17:50:17 -0500

Seen: 2,105 times

Last updated: Jan 27 '18