Multi-sensor fusion positioning (1-3D laser odometer) 6-realize the accuracy evaluation of calling pcl-icp, ndt -3 evo odometer
Reference blog:
Autopilot positioning from scratch (7): Accuracy evaluation of odometer
Download evo tool
pip install evo --upgrade --no-binary evo
evo usage, taken from teacher Ren's knowledge
EVO evaluation data has two modes, the corresponding commands are evo_rpe and evo_ape respectively. The former evaluates the error within each distance, and the latter evaluates the accumulation of absolute error with distance.
To evaluate the error within each distance, you can use the following commands
evo_rpe kitti ground_truth.txt laser_odom.txt -r trans_part --delta 100 --plot --plot_mode xyz
Among them, -delta 100 means that the error is counted every 100 meters, so that the statistics are actually the percentage of the error, which can directly correspond to the distance error index in the odometry list of kitti.
Perform the ICP odometer registration in the previous chapter , and you can see the ground_truth.txt and laser_odom.txt files in the slam_data/trajectory directory, which are GNSS (think true value) and post-ICP odometer (estimated pose)
Result evaluation:
ICP
ICP RVIZ visualized trajectory graph
evo_rpe kitti ground_truth.txt laser_odom.txt -r trans_part --delta 100 --plot --plot_mode xyz
Maximum error: 5908.669974%
Average error: 711.311504% Median error:
105.818037%
Minimum error: 0.828498%
RMSE: 1504.886112%
The following command can be used to evaluate the total cumulative error
evo_ape kitti ground_truth.txt laser_odom.txt -r full --plot --plot_mode xyz
Maximum value: 9207.618971
Average value: 1064.460571
RMSE: 1943.251093
The gray part is the cumulative error
NDT
NDT RVIZ visualized trajectory chart
evo_rpe kitti ground_truth.txt laser_odom.txt -r trans_part --delta 100 --plot --plot_mode xyz
Maximum error: 0.281753%
Average error: 0.036494% Median error:
0.036494%
Minimum error: 0.036494%
RMSE: 0.036494%
The following command can be used to evaluate the total cumulative error
evo_ape kitti ground_truth.txt laser_odom.txt -r full --plot --plot_mode xyz
