SkillAgentSearch skills...

NDTMC

[IROS 2024] A 3D Global Descriptor For Loop Closure Detection. NDT-Map-Code.

Install / Use

/learn @SlamCabbage/NDTMC
About this skill

Quality Score

0/100

Supported Platforms

Universal

README

NDT-Map-Code

NDT-Map-Code: A 3D global descriptor for real-time loop closure detection in lidar SLAM

You can find the paper here NDT-Map-Code.

News <a name="news"></a>

  • [2024/2]:fire: NDT-Map-Code is accepted by IROS 2024.

Demo

NDT-Map-Code

Example

  • We implemented the SLAM package of NDTMC and LIOSAM integration, which can be found at NDTMC-LIO-SAM.

  • We tested our descriptor on KITTI sequences 00, 02, 05, 06, 07 and 08:

<p align="center"><img width="800" alt="image" src="script/000205.png"> <p align="center"><img width="800" alt="image" src="script/060708.png">
  • F1 score and extended precision results on the KITTI dataset:
<p align="center"><img width="800" alt="image" src="script/Table2.png">

How to use

  1. Create GT of LCD

    Use script/create_ground_truth.py to generate the LCD GT value, you need to modify the two file paths in the script/create_ground_truth.py (odometry ground truth file path, LCD GT value result path).

  2. Save descriptors

    Modify the main function in run_demo.cpp as follow:

    int main(int argc, char **argv)
    {
        saveDesc(argv);
        // matchForKitti(argv);
        // matchForNIO(argv);
        return 0;
    }
    
    

    Build and run script/auto_test.py to generate descriptors, where input_folder and output_folder need to be modified to the voledyne folder path of the currently processed sequence and the folder path where you want to save the descriptor results, respectively.

  3. Descriptors matching

    Modify the main function in run_demo.cpp as follow:

    int main(int argc, char **argv)
    {
        // saveDesc(argv);
        matchForKitti(argv);
        // matchForNIO(argv);
        return 0;
    }
    
    

    Build and run script/auto_match.py to perform descriptor matching, where input_folder and output_txt need to be modified to the folder path where the descriptor was saved in the previous step and the txt file path where you want to save the matching result, respectively.

At this point, you can get the matching result of each frame, and you can process it yourself to get more result information, such as: precision-recall curve, F1 Score and Extended Precision.

Cite NDTMC

@inproceedings{liao2024ndt,
  title={NDT-Map-Code: A 3D global descriptor for real-time loop closure detection in lidar SLAM},
  author={Liao, Lizhou and Yan, Wenlei and Sun, Li and Bai, Xinhui and You, Zhenxing and Yuan, Hongyuan and Fu, Chunyun},
  booktitle={2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  pages={7854--7861},
  year={2024},
  organization={IEEE}
}

Contact

  • Maintainer: Lizhou Liao (liaolizhou@icloud.com)

Contributors

  • Lizhou Liao: completed the code

Acknowledgement

  • Thanks for NIO low-speed localization and mapping group.
View on GitHub
GitHub Stars207
CategoryDevelopment
Updated3d ago
Forks11

Languages

C++

Security Score

95/100

Audited on Apr 3, 2026

No findings