Trajectory Error of Visual Odometry(RGB-D node)

[takahiko-hasegawa-hdjp]

I am currently trying to evaluate the trajectory error of visual odometry using rgb-d node.

I found out that the visual odometry node isnt performing very well when the baselink turns to the right. The camera is rotated 90 degress roughly but the comapring that to ground truth the trajectory is a bit off.

What can I do to make the trajectory better?

I understatnd I may need to do a bit more research on improving the odometry result but I was wondering if there is anything that could be done to VO config to improve the trajectory?

Thanks!

[matlabbe]

Well, this is odometry drift, looks like it didn't estimate perfectly the 90 deg turn. There are many things that can affect odometry accuracy:

• Calibration (depth accuracy? distortions? depth registration? ...)
• Environment (enough features? ...)
• Camera quality (motion blur while rotating? global vs rolling shutter? ...)

I suggest to record and share a rosbag with all topics if you want more specific help about your case.