The official implementation of CVAT (A Curved Voxel Association Triangle Descriptor for Universal LiDAR Place Recognition), a universal LiDAR place recognition method. CVAT is a robust descriptor combining curved-surface feature and triangle feature. Through FOV Alignment strategy and fine triangle verification, we achieve universal LiDAR place recognition and accuracy 6-DOF pose estimation.

Welcome to our website for more details.

This project was tested on Ubuntu 20.04 and ROS Noetic.

Make sure the following dependencies are installed:

• CMake >= 3.20
• PCL >= 1.8.0
• Eigen >= 3.2.1

Clone this repository and build:

mkdir cvat_ws/src -p
cd cvat_ws/src
git clone https://github.com/Yanpiii/CVAT.git
cd .. && catkin_make
source ~/cvat_ws/src/devel/setup.bash

KITTI/KITTI360

Modify the lidar_path and pose_path in the launch file. Note that for KITTI and KITTI360, when loading point clouds in the .bin format and poses in the tum format (timestamp x y z qx qy qz qw), it is recommended to refer to the format of Semantic KITTI.

roslaunch cvat_descriptor cvat_kitti.launch

HeLiPR

Modify the lidar_path, pose_path, and lidar_type (Ouster, Velodyne, Avia, Aeva) in the launch file.

roslaunch cvat_descriptor cvat_helipr.launch

MCD

Modify the bag_path in the launch file. Note that for the MCD, we create a .bag file with submaps and localization groundtruth based on the official provided .pcd files. The demo data can be found here. We just test Livox Mid 70 LiDAR.

roslaunch cvat_descriptor cvat_mcd.launch

We test cross-LiDAR place recognition on the HeLiPR dataset. Modify the src_lidar_path, tgt_lidar_path, src_pose_path, and tgt_pose_path in the launch file.

And, you should modify src_type and tgt_type (Ouster, Velodyne, Avia, Aeva) in the launch file for achieving "from source lidar to target lidar place recognition".

roslaunch cvat_descriptor cvat_multi_lidar.launch

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.