search

rpng / open_vins

An open source platform for visual-inertial navigation research.
https://docs.openvins.com
GNU General Public License v3.0
2.19k stars 647 forks source link

IMU calibration model to improvements? #68

Closed Wangxuefeng92 closed 4 years ago

Wangxuefeng92 commented 4 years ago

How about change IMU calibration model to improve the precision of VIO?

WoosikLee2510 commented 4 years ago

Can you suggest what model you have in your mind? Is it IMU intrinsic or extrinsic? We recently published IMU intrinsic calibration method (Analytic Combined IMU Integration for Visual-Inertial Navigation, ICRA2020) so this might be what we will add in our system in the future.

yangyulin commented 4 years ago

Thanks for your interest in OpenVINS. Our colleagues have given you some feedback regarding your questions above. I will just give some additional comments.

1) Regarding the IMU calibration model, do you refer to the IMU noise model (like how to model biases) or IMU intrinsics model?

2) If for the IMU noise model, OKVIS used a slightly different bias model (at least from their paper). I have no experience on that bias model yet, but I guess it won't change the accuracy much. http://www.roboticsproceedings.org/rss09/p37.pdf

3) If for the IMU intrinsic model, we have a recent paper on adding IMU intrinsic calibration to OpenVINS with 4 model variants. It can improve the system performance only when you can fully excite the 6D motion of the IMU. We find out that the IMU intrinsic calibration is very sensitive to motions. The paper will come out soon in July @ RSS2020.

On Thu, May 21, 2020 at 11:53 AM Woosik Lee notifications@github.com wrote:

Can you suggest what model you have in your mind? Is it IMU intrinsic or extrinsic? We recently published IMU intrinsic calibration method (Analytic Combined IMU Integration for Visual-Inertial Navigation, ICRA2020) so this might be what we will add in our system in the future.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/rpng/open_vins/issues/68#issuecomment-632171726, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADJAQCACEEQULED3VMFI7DLRSVE7XANCNFSM4NGPAJNQ .

-- With Best Regards, Linde Yang / Yulin Yang / 杨玉林

UNIVERSITY of DELAWARE PhD Candidate of Mechanical Engineering 328 Spencer Lab, Newark, DE, USA, 19716-3140 Mobile: +(01) 302-690-3926 mailto: yyl2038@163.com yuyang@udel.edu Web: http://udel.edu/~yuyang/ http://udel.edu/~yuyang/

Wangxuefeng92 commented 4 years ago

thank you for your replies!