SangwooJung98 / 3D_Map_Evaluation

Quantitative 3D Map Accuracy Evaluation Hardware and Algorithm for LiDAR(-Inertial) SLAM [IEEE ICCAS 2024]

3D_Map_Accuracy_Evaluation

Quantitative 3D Map Accuracy Evaluation Hardware and Algorithm for LiDAR(-Inertial) SLAM

Authors

Sanghyun Hahn^†, Seunghun Oh^†, Minwoo Jung, Ayoung Kim and Sangwoo Jung^*.

Paper

This repository includes the Quantitative 3D Map Accuracy Evaluation Hardware and Algorithm for LiDAR(-Inertial) SLAM, which is accepted to IEEE ICCAS 2024.

[arXiv] [BibTex]

Setup

python 3.11

pip install -r requirements.txt

Cropping

target_manual.py

This process can be tested with our demo pointcloud, original_pointcloud.pcd.

Enter the GPS pose that you wish to crop around into target_pos : the code will automatically provide and visualize the loosely cropped pointcloud.

Then, crop the target tightly manually to obtain the tightly cropped pointcloud.

vid.mp4

keyboard instructions for cropping

X Y Z: axis view  
F: free look  
Mouse left drag : view adjustment  
Mouse wheel up/down : zoom in/out  
K: enter/leave cropping mode   
Ctrl + Mouse left click : select region (in cropping mode)  
Mouse left drag : select region (in cropping mode)  
C: crop  
S: save **add .pcl for filename extension **   

Target pose estimation

target_pose_est.py

This process can be tested with our demo tightly cropped pointcloud, cropped_pointcloud.ply.

K-means Clustering, RANSAC, SVD is applied to the tightly cropped pointcloud.

As a result, the script returns the x, y position of the target. (=estimated target pose)

While running, # of iterations and total attempts are visualized.

Error Calculation

Prepare the ground truth pose and estimated target pose for each target.

Ground Truth

The ground truth can be obtained from the GPS sensor attatched to the target.

Subtract the rtk origin pose from the rtk target pose for the ground truth pose.

Copy and paste each pose into relative_absolute.py

Error Metrics

relative_absolute.py returns

1. frame translation/rotation values
2. Distance errors between corresponding targets
3. Absolute error (the average of 2)
4. Errors of each segment
5. Relative error (the average of 4)

Contact

This repository is provided for academic purposes. If you encounter technical problems, please contact <Sanghyun Hahn: [e-mail]>, <Seunghun Oh: [e-mail]>, or <Sangwoo Jung: [e-mail]>.

BibTex

@article{hahn2024quantitative,
  title={Quantitative 3D Map Accuracy Evaluation Hardware and Algorithm for LiDAR (-Inertial) SLAM},
  author={Hahn, Sanghyun and Oh, Seunghun and Jung, Minwoo and Kim, Ayoung and Jung, Sangwoo},
  journal={arXiv preprint arXiv:2408.09727},
  year={2024}
}