search

MichaelGrupp / evo

Python package for the evaluation of odometry and SLAM
https://michaelgrupp.github.io/evo/
GNU General Public License v3.0
3.43k stars 746 forks source link

Questions about Umeyama Alignment and Evaluation Results #599

Closed wangjinhoon closed 10 months ago

wangjinhoon commented 11 months ago

Hello, I wonder why the results of the two statements are different. Can't evo_ape also get the right traj like evo_traj??

evo_traj tum 1030test31.14.txt --ref 1030test32.5.txt --plot

Untitled

evo_ape tum 1030test32.5.txt 1030test31.14.txt --plot

Untitled (1)

Also, the path has been changed, and I wonder if this is the correct error value.

thank!

MichaelGrupp commented 11 months ago

evo_ape by default shows only the synchronized poses that were used for the metrics calculation. But you can show the "full" reference with --plot_full_ref, i.e. including poses that were not used by the metric.

evo_traj shows the full reference because it can align multiple trajectories.

wangjinhoon commented 11 months ago

Thank you for your quick reply.

Does synchronization mean that Roll, Pitch, and Yaw are the same on the same timestamp when aligned?

I want to compare the Vive tracker with the VO. Is there anything I should care about when creating test data to increase the number of synchronized poses? Currently, my data is very different in the amount of data and does not take into account roll, pitch, or yaw.

I'm not sure if the values obtained with a small number of synchronized poses are correct.

thank

MichaelGrupp commented 10 months ago

Umeyama alignment rigidly transforms (optionally with scale) your trajectory so that it fits the best way possible to the other one. This doesn't mean that the RPY angles are the same afterwards.

wangjinhoon commented 10 months ago

I increased the data volume and resolved the issue. Thank you