1. Github address
2. Install evo
3. Evo command line interactive
Metrics:
evo_ape-absolute attitude error
evo_rpe-relative attitude error
Tools:
evo_traj-analyze, draw, and output one or more trajectories
evo_res-compare one or more result files of
evo_ape /eco_rpe evo_fig-reopen the serialized image (stored as serialize_plot)
evo_config-global settings and configuration file operations
1) To draw multiple trajectories, evo_traj
supports formats such as tum, kitti, and eurco. Use ORB_SLAM2 to run the TUM data set and get two txt files, use evo to display them together with groundtruth
evo_traj tum CameraTrajectory.txt KeyFrameTrajectory.txt groundtruth.txt -p --plot_mode=xyz
Result:
But groundtruth is far away from these two tracks?
There are the following parameters
-p or -plot plotting
-v or -verbose output related information (mean, variance, etc.)
-f or -full_check check related information (whether the timestamp corresponds to whether the quaternion is a unit quaternion)
-a or –Align to register the trajectory, using the ICP method, not just to align the starting point. The
application of the alignment requires a given –ref
--align or -a = SE(3) Umeyama alignment (rotation, translation)
--align --correct_scale or -as = Sim(3) Umeyama alignment (rotation, translation, scale)
--correct_scale or -s = scale correction only
For example:
before the camera trajectory is registered with the groundtruth
evo_traj tum CameraTrajectory.txt groundtruth.txt -p
After registration
evo_traj tum CameraTrajectory.txt --ref=groundtruth.txt -p -a
2) Measure
evo_ape on the trajectory : Absolute pose error, often used as absolute trajectory error, compare estimated trajectory and reference trajectory and calculate the statistical data of the entire trajectory, suitable for testing the global consistency of the trajectory.
Command syntax: The command format refers to the trajectory estimation trajectory [optional]. The
optional options include alignment commands, drawing, and saving results.
evo_ape tum groundtruth.txt CameraTrajectory.txt -r full -va -p --save_results results/ORB.zip
Where -r represents the attitude relationship based on ape
-v represents verbose mode, the detailed mode
can be viewed through evo_ape tum --help to view the parameter
evo_rpe : relative pose error compare motion (posture increment), relative pose error can give local accuracy , Such as the translation or rotation drift of the slam system per meter.
Command syntax: The command format refers to the trajectory estimation trajectory [optional]. The
optional options include alignment commands, drawing, and saving results.
Reference link: SLAM accuracy evaluation-EVO
evo official link wiki