HKUST-Aerial-Robotics/Pagor

This is the official code repository of "Pyramid Semantic Graph-based Global Point Cloud Registration with Low Overlap", which is accepted by IROS'23.

Pagor (PyrAmid Graph-based GlObal Registration) is a robust global point cloud registration algorithm for LiDAR. It takes two point clouds and their semantic labels as input and estimates the relative pose between them.

• An improved version can be found in G3Reg.
• Welcome to try our new LiDAR Registration Benchmark which is a comprehensive benchmark for LiDAR registration in robotic applications.

Follow the official guide to install ROS1.

Follow the official guide to install GTSAM

Follow the official guide to install PCL

sudo apt install cmake libeigen3-dev libboost-all-dev libgoogle-glog-dev libyaml-cpp-dev

mkdir -p catkin_ws/src
cd catkin_ws/src
git clone https://github.com/HKUST-Aerial-Robotics/Pagor.git
cd .. && catkin_make

If you want to reproduce the KITTI benchmark results of the paper, you can download the KITTI semantic labels [OneDrive][Baidu Cloud], which is generated by LiDAR segmentation model SalsaNext. Then merge the downloaded folder with the original KITTI odometry LiDAR dataset, and then modify the path in the configuration file configs/pagor.yaml.

dataset:
  name: kitti
  dataset_path: "/media/qzj/Document/datasets/KITTI/odometry/data_odometry_velodyne/dataset/sequences"
  split_dir: "data_split/kitti"
  label_dir: "/labels_salsanext/"

The dataset folder structure is as follows:

dataset/
└── sequences
    └── 00
        ├── calib.txt
        ├── labels_salsanext
        ├── poses.txt
        ├── times.txt
        └── velodyne

Run the following command:

source devel/setup.bash
cd src/Pagor
../../devel/lib/pagor/kitti_bm pagor.yaml

If you find this work useful in your research, please consider citing:

@inproceedings{qiao2023pyramid,
  title={Pyramid Semantic Graph-based Global Point Cloud Registration with Low Overlap},
  author={Qiao, Zhijian and Yu, Zehuan and Yin, Huan and Shen, Shaojie},
  booktitle={2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  pages={11202--11209},
  year={2023},
  organization={IEEE}
}

This project is licensed under the MIT License - see the LICENSE file for details.

We want to express our deepest gratitude to the creators of the repositories listed below for sharing their work with the public:

• Teaser (baseline)
• Quatro (baseline)
• Segregator (baseline)
• CLIPPER (maximum clique solver)
• T-LOAM (clustering)
• SalsaNext (semantic segmentation)