search

ethz-asl / rovio

Other
1.14k stars 507 forks source link

Deviation in trajectory #219

Open rocomanmelli opened 5 years ago

rocomanmelli commented 5 years ago

Hi!

We are running your visual inertial odometry system on a public agricultural dataset recorded by us (you can check it out here: http://www.cifasis-conicet.gov.ar/robot/doku.php). Some of the results we got are shown in this screenshot:

rovio_issue

As you can see, the trajectories estimated by Rovio, except in sequence 1, tend to have some kind of deviation with respect to the ground truth. The configuration files used are:

zavalla_cam0.txt zavalla_cam1.txt rovio_zavalla_sequence01.txt -> For sequences 1, 3 and 4. rovio_zavalla_sequence05.txt -> For sequence 5.

Intrinsic cameras calibrations and extrinsic imu-camera calibrations were obtained with Kalibr and other SLAM systems (SPTAM, ORB_SLAM2, MSCKF_VIO) were tested with them on this dataset and we had no problems. However we do have some doubts about the covariances and other configurations.

One thing we know that could be wrong is the synchronization between sensors. However, we would expect the whole system to crash from the start if that was the main problem.

Is there any recommendation you could give us? Do you have any idea of what could be happening?

Thanks in advance.

Román.

mhkabir commented 5 years ago

Your IMU and camera are not synchronised. As Rovio is a tightly coupled visual-inertial system, you cannot expect any results unless you have accurate timestamping of camera images and IMU data against the same clock.