Ask Your Question
0

Filtering Visual Odometry and IMU/GPS data

asked 2014-09-16 04:24:10 -0600

mister_kay gravatar image

updated 2014-09-16 04:24:53 -0600

Hello,

during the past months I have been working on gettings results from a stereo-cam system using ROS-package viso2_ros in a moving vehicle. So far so good, the egomotion estimation from visual odometry really works well as can be seen here:

https://docs.google.com/document/d/1l...

I now have wheel_odometry, visual_odometry, IMU and GPS-data in ASCII-files ready to be processed and filtered (also have a reference trajectory giving the "true" position). Here is the data:

https://drive.google.com/file/d/0B1b1...

I can filter all the data with the package "robot_pose_ekf". The disadvantage is that the package uses the orientation from the IMU rather than angular rates and acceleration. This orientation is drifting and not accurate.

I was wondering if you could give me tips how to build a filter which fuses all the described data. If possible the design of the filter should not be that complicated since I donot have that much knowledge about EKFs. Also any good literature recommendation is appreciated.

Hector

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted
1

answered 2014-09-16 08:11:21 -0600

Tom Moore gravatar image

Have you tried robot_localization? It accepts angular velocities, though linear acceleration is not (yet) supported.

However, I'm surprised that you're getting drift from robot_pose_ekf. Have you tried plotting the raw sensor data?

edit flag offensive delete link more

Comments

Yes I did. Actually that is what my supervisor told me, e.g. if an xsens is in rest, the orientation [computed by xsens] will drift over time. I also have some experience with that. My former colleague has put up a video of his error state filter: https://www.youtube.com/watch?v=ymuhJ... .

mister_kay gravatar imagemister_kay ( 2014-09-16 09:59:27 -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

3 followers

Stats

Asked: 2014-09-16 04:24:10 -0600

Seen: 1,370 times

Last updated: Sep 16 '14