uzh-rpg/rpg_trajectory_evaluation

One question about visual-inertial case?

Opened this issue · 1 comments

democheng commented

Hi, I have read the paper A Tutorial on Quantitative Trajectory Evaluation for Visual(-Inertial) Odometry, which is a great paper with useful tool written in Python.

But I have one question about visual-inertial case:
the paper recommends to use a 4-DOF transform, but the groundtruth may not be provided in world frame(e.g.,VICON), should I transform the groundtruth to world frame? Or it would be all right to use a 6-DOF transform? How different is it?

zhangzichao commented

Hi, sorry for the late reply. If you think there is a non-negligible error regarding the gravity direction in the groundtruth, there is not much that can be done to reason properly about roll and pitch. And if the trajectory is aligned with a 6 DoF transformation in this case, I think at least the position error can still be of some value.