Page 1 of 1

task 1 reference frame definition

PostPosted: Mon 4. Aug 2014, 15:31
by unipi.euroc.team
task 1 requires to provide imu pose:

Response (geometry_msgs/PoseStamped)
Contains the pose of the IMU (!). Use the timestamp and sequence information contained in
the header of image[0], and write this into the header of the pose.

How is the camera frame positioned on the camera with respect to
the optical axis and horizontal direction?
Should we assume the convention (X Right, Y Down, Z out). ?

Which reference frame should we use as "world" for imu pose ?
in other words: the imu pose must be computed with respect to a certain fixed/inertial reference frame.
Which frame should we use?

Something similar to what we see in section 3.5.2 of the Challenge 3 ‒ Technical Annex for
Stage I pdf document?

Thsnks,
Team UNIPI

Posted: Mon 4. Aug 2014, 15:31
by Advertising

Re: task 1 reference frame definition

PostPosted: Tue 5. Aug 2014, 17:07
by ffurrer
For the camera we follow the convention you suggested
    x-right, y-down, z-out
in general that is the openCV convention.

You can pick any inertial ("world") frame, only relative errors are punished in the evaluation. This is described in detail on the KITTI framework website http://www.cvlibs.net/datasets/kitti/ as referred to in the technical annex.


Cheers
Fadri