Point Cloud Structural Similarity-based Underwater Sonar Loop Detection

[Video 📺], [Paper 📖]

Donghwi Jung, Andres Pulido, Jane Shin, Seong-Woo Kim

This is an implentation of "Point cloud structural similarity-based underwater sonar loop detection" which indicates detecting loops based on the structural similarity of point clouds generated from the data acquired by MBES.

System Overview

Dependencies

• The script generate_data_from_antarctica.py relies on the AUVLib library.
  • To run this script, follow the setup instructions provided in the AUVLib repository.
  • All other scripts in this project are independent of AUVLib.

Installation

conda env create -f environment.yaml
conda activate PCSS

Dataset

Download

• Download datasets
• Unzip the downloaded dataset

Data processing

If you would like to generate the data yourself, follow the process outlined below. To process the Antarctica dataset, the auv_lib library must be installed. The antarctica_2019.cereal file is also required. You can download the file.

python generate_data_from_antarctica.py

or

python generate_data_from_seaward.py
# ex) python  generate_data_from_seaward.py --data_id 3

Execution

python execute.py
# ex) python execute.py --data_path data --data_id 1 --neighborhood_size 100 --score_threshold 2.95

Acknowledgements

The codes and datasets in this repository are based on PointSSIM, Antarctica, and Seaward. Thanks to the authors of these codes and datasets.

Citation

@article{jung2025point,
  title={Point Cloud Structural Similarity-Based Underwater Sonar Loop Detection},
  author={Jung, Donghwi and Pulido, Andres and Shin, Jane and Kim, Seong-Woo},
  journal={IEEE Robotics and Automation Letters},
  year={2025},
  publisher={IEEE}
}