IFTD: Image Feature Triangle Descriptor for Loop Detection in Driving Scenes

1. Introduction

IFTD_detector In this work, we propose a fast and robust Image Feature Triangle Descriptor (IFTD) based on the STD method, aimed at improving the efficiency and accuracy of place recognition in driving scenarios. We extract keypoints from BEV projection image of point cloud and construct these keypoints into triangle descriptors. By matching these feature triangles, we achieved precise place recognition and calculated the 4-DOF pose estimation between two keyframes. Furthermore, we employ image similarity inspection to perform the final place recognition. Experimental results on three public datasets demonstrate that our IFTD can achieve greater robustness and accuracy than state-of-the-art methods with low computational overhead.

1.1. Developers:

The codes of this repo are contributed by: Fengtian Lang, Ruiye Ming, Zikang Yuan.

1.2. Related paper

Our preprint version is now available on arxiv:
IFTD: Image Feature Triangle Descriptor for Loop Detection in Driving Scenes

2. Prerequisites

2.1 Requirements

GCC >= 9.4.0

Cmake >= 3.11.0

Eigen3 >= 3.3.4

Ceres >= 2.1

OpenCV >= 3.4

PCL == 1.8 for Ubuntu 18.04, and == 1.10 for Ubuntu 20.04

ROS

Have Tested On:

OS	GCC	Cmake	Eigen3	OpenCV	PCL
Ubuntu 18.04	9.4.0	3.11.2	3.3.4	3.4.16	1.8.0

2.2 Create ROS workspace

mkdir -p ~/IFTD/src
cd IFTD/src

2.3 Clone the directory and build

git clone https://github.com/EinsTian1/iftd.git
cd ..
catkin_make

2.4 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. 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 Mulran 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 Mulran Dataset website.
For the Nclt 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 NCLT Dataset website.

2.5.2. 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. Run on Public Datasets

Noted:

Please create a folder named "output" in "xxxx" folder before running. When IFTD is running, the result of loop closure is recorded in real time in the loop.txt located in the output folder.

3.1 Run on KITTI

First, you should modify the demo_kitti.launch file

Set the lidar_path to your local path
Set the pose_path to your local path

Then go to the workspace of IFTD and open a terminals, type the following command in terminal :

cd IFTD
source devel/setup.bash
roslaunch iftd_detection demo_kitti.launch

3.2 Run on Mulran

First, you should modify the demo_mulran.launch file

Set the lidar_path to your local path
Set the pose_path to your local path

Then go to the workspace of IFTD and open a terminals, type the following command in terminal :

cd IFTD
source devel/setup.bash
roslaunch iftd_detection demo_mulran.launch

3.3 Run on NCLT

First, you should modify the demo_nclt.launch file

Set the lidar_path to your local path
Set the pose_path to your local path

Then go to the workspace of IFTD and open a terminals, type the following command in terminal :

cd IFTD
source devel/setup.bash
roslaunch iftd_detection demo_nclt.launch

Citation

If you use our work in your research project, please consider citing:

@misc{lang2024iftd,
      title={IFTD: Image Feature Triangle Descriptor for Loop Detection in Driving Scenes}, 
      author={Fengtian Lang and Ruiye Ming and Zikang Yuan and Xin Yang},
      year={2024},
      eprint={2406.07937},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Acknowledgments

Thanks for STD.