Skip to content

Latest commit

 

History

History
172 lines (136 loc) · 10.5 KB

README.md

File metadata and controls

172 lines (136 loc) · 10.5 KB
Raw

STD: A Stable Triangle Descriptor for 3D place recognition

1. Introduction

STD_detector is a global descriptor for 3D place recognition. For a triangle, its shape is uniquely determined by the length of the sides or included angles. Moreover, the shape of triangles is completely invariant to rigid transformations. Based on this property, we first design an algorithm to efficiently extract local key points from the 3D point cloud and encode these key points into triangular descriptors. Then, place recognition is achieved by matching the side lengths (and some other information) of the descriptors between point clouds. The point correspondence obtained from the descriptor matching pair can be further used in geometric verification, which greatly improves the accuracy of place recognition.

A typical place recognition case with STD. These two frames of point clouds are collected by a small FOV LiDAR (Livox Avia) moving in opposite directions, resulting in a low point cloud overlap and drastic viewpoint change.

1.1. Developers:

The codes of this repo are contributed by: Chongjian Yuan (袁崇健), Jiarong Lin (林家荣) and dustier

1.2. Related paper

Our paper has been accepted to ICRA2023, and our preprint version is now available on arxiv:
STD: Stable Triangle Descriptor for 3D place recognition

1.3. Related video

Our accompanying video is now available on YouTube.

2. Prerequisites

2.1 Ubuntu and ROS

We tested our code on Ubuntu18.04 with ros melodic and Ubuntu20.04 with noetic. Additional ROS package is required:

sudo apt-get install ros-xxx-pcl-conversions

2.2 Eigen

Following the official Eigen installation, or directly install Eigen by:

sudo apt-get install libeigen3-dev

2.3. ceres-solver (version>=2.1)

Please kindly install ceres-solver by following the guide on ceres Installation. Notice that the version of ceres-solver should higher than ceres-solver 2.1.0

2.4. GTSAM

Following the official GTSAM installation, or directly install GTSAM 4.x stable release by:

# Add PPA
sudo add-apt-repository ppa:borglab/gtsam-release-4.0
sudo apt update  # not necessary since Bionic
# Install:
sudo apt install libgtsam-dev libgtsam-unstable-dev

!! IMPORTANT !!: Please do not install the GTSAM of develop branch, which are not compatible with our code! We are still figuring out this issue.

2.5 Prepare for the data

Since this repo does not implement any method (i.e., LOAM, LIO, etc) for solving the pose for registering the LiDAR scan. So, you have to prepare two set of data for reproducing our results, include: 1) the LiDAR point cloud data. 2) the point cloud registration pose.

2.5.1. Download our Example data

Departure from the purpose of convenience, we provide two sets of data for your fast evaluation, which can be downloaded from OneDrive and BaiduNetDisk(百度网盘)

2.5.2. LiDAR Point cloud data

  • For the Kitti dataset (i.e., our Example-1), we read the raw scan data with suffix ".bin". These raw LiDAR scan data can be downloaded from the Kitti Odometry benchmark website.
  • For the solid-state LiDAR dataset (i.e., our Example-2), we read the undistort scan data from the recorded rosbag files, whose bag file contains undistort LiDAR scan data in rostopic: "/cloud_undistort"

2.5.3. Point cloud registration pose

In the poses file, the poses for LiDAR point cloud registration are given in the following data format:

Timestamp pos_x pos_y pos_z quat_x quat_y quat_z quat_w

where, Timestamp is the correspond sampling time stamp of a LiDAR scan, pose_{x,y,z} and quad_{x,y,z,w} are the translation and rotation (expressed used quaternion) of pose.

3. Examples

This reposity contains implementations of Stable Triangle Descriptor, as well as demos for place recognition and loop closure correction. For the complete pipline of online LiDAR SLAM, we will release this code along with the release of the extended version.

3.1. Example-1: place recognition with KITTI Odometry dataset

To run Example-1, you need to first download the poses file we provide.

Then, you should modify the demo_kitti.launch file

  • Set the lidar_path to your local path
  • Set the pose_path to your local path
cd $STD_ROS_DIR
source deve/setup.bash
roslaunch std_detector demo_kitti.launch

3.2. Example-2: place recognition with Livox LiDAR dataset

To run Example-2, you need to first download the rosbag file and poses file we provide. Then, you should modify the demo_livox.launch file

  • Set the bag_path to your local path
  • Set the pose_path to your local path
cd $STD_ROS_DIR
source deve/setup.bash
roslaunch std_detector demo_livox.launch

3.3. Example-3: loop closure correction on the KITTI Odometry dataset

The point cloud map and trajectory before and after correction by STD.

To run Example-3, you need to first download the poses file we provide or create your own pose file on the KITTI Odometry dataset with a LiDAR odom following the format: timestamp x y z qx qy qz qw Then, you should modify the demo_pgo.launch file

  • Set the lidar_path to your local path
  • Set the pose_path to your local path
cd $STD_ROS_DIR
source deve/setup.bash
roslaunch std_detector demo_pgo.launch

3.4. Example-4: online loop closure correction with FAST-LIO2 integrated

To run Example-4, you need to install and configure FAST-LIO2 first. You can try the data building_slower_motino_avia.bag here(provided by zlwang7), which is outdoor scan data with no loop closure other than the one between the starting point and the endpoint. Therefore, relying solely on the fast-lio algorithm results in obvious Z-axis drift, with STD loop detection and graph optimization, there will be a noticeable correction to the drift.

# termianl 1: run FAST-LIO2
roslaunch fast_lio mapping_avia.launch

# terminal 2: run std online demo
roslaunch std_detector demo_online.launch

# terminal 3: play data
rosbag play building_slower_motion_avia.bag

Acknowledgments

In the development of STD_detector, we stand on the shoulders of the following repositories:

  • Scan Context: An Egocentric Spatial Descriptor for Place Recognition within {3D} Point Cloud Map
  • FAST-LIO: A computationally efficient and robust LiDAR-inertial odometry package.
  • VoxelMap: An efficient and probabilistic adaptive(coarse-to-fine) voxel mapping method for 3D LiDAR.
  • R3LIVE: A Robust, Real-time, RGB-colored, LiDAR-Inertial-Visual tightly-coupled state Estimation and mapping package

Contact Us

We are still working on improving the performance and reliability of our codes. For any technical issues, please contact us via email Chongjian Yuan < ycj1ATconnect.hku.hk >, Jiarong Lin < ziv.lin.ljrATgmail.com >.

For commercial use, please contact Dr. Fu Zhang < [email protected] >

License

The source code of this package is released under GPLv2 license. We only allow it free for personal and academic usage. For commercial use, please contact us to negotiate a different license.

We are still working on improving the performance and reliability of our codes. For any technical issues, please contact contact us via email Chongjian Yuan < ycj1ATconnect.hku.hk >, Jiarong Lin < ziv.lin.ljrATgmail.com >.

If you use any code of this repo in your academic research, please cite at least one of our papers:

[1] Yuan, C., Lin, J., Zou, Z., Hong, X., & Zhang, F.. "STD: Stable Triangle Descriptor for 3D place recognition."
[2] Xu, W., Cai, Y., He, D., Lin, J., & Zhang, F. "Fast-lio2: Fast direct lidar-inertial odometry."
[3] Yuan, C., Xu, W., Liu, X., Hong, X., & Zhang, F. "Efficient and probabilistic adaptive voxel mapping for accurate online lidar odometry."