FGO-MOT

This repository is a core 3D MOT component within SLAMMOT. It performs continuous object tracking and trajectory estimation for robust online tracking and smooth object trajectories.

KITTI Demo1

KITTI Demo2

Highlights

• Factor-graph optimization: models motion priors, observation constraints, and marginalization as factors and solves them in a unified optimization framework.

Common Dependencies

• ROS 1 (Noetic recommended)
• PCL, Eigen, and OpenCV
• Ceres Solver

Quick Start

1. Prepare a catkin workspace and place this package in src/:

mkdir -p ~/catkin_ws/src
cd ~/catkin_ws/src
git clone https://github.com/GREAT-WHU/FGO-MOT.git
cd ..
catkin_make
source devel/setup.bash

2. Prepare the KITTI point clouds and images. Calibration, pose, OXTS, and PointRCNN detection files are already provided under data/.

kitti_helper:
  base_path: "data/"
  detection_folder: "pointrcnn_Car_val"

3. Launch tracking:

roslaunch fgo_mot run.launch

Run without RViz if needed:

roslaunch fgo_mot run.launch visualization:=false

Data and Configuration

• Dataset:

  • Calibration, pose, OXTS, and PointRCNN detection files are included under data/.
  • Download KITTI Tracking and copy only the point clouds and stereo images into this directory.

  The completed data directory should be:

  data/
  ├── calib/                 # provided
  ├── oxts/                  # provided
  ├── pose/                  # provided
  ├── pointrcnn_Car_val/     # provided detections
  ├── velodyne/              # copy from KITTI tracking/training
  │   ├── 0000/
  │   ├── ...
  │   └── 0020/
  ├── image_02/              # copy from KITTI tracking/training
  │   ├── 0000/
  │   ├── ...
  │   └── 0020/
  └── image_03/              # copy from KITTI tracking/training
      ├── 0000/
      ├── ...
      └── 0020/
  

• Detection directory:

  detection_folder is relative to base_path. For example:

  detection_folder: "pointrcnn_Car_val"

  reads detections from:

  <base_path>/pointrcnn_Car_val/0001.txt
  

• Parameters:

  • Tracking and optimization parameters: config/config.yaml.

• Evaluation:

  • Results are saved under output/tracking_kitti/ relative to the package directory.
  • Evaluate with the official KITTI Tracking tools or TrackEval for CLEAR MOT, HOTA, and identity metrics.

Validation Results

The following results were evaluated on the KITTI Tracking validation sequences 0001, 0006, 0008, 0010, 0012, 0013, 0014, 0015, 0016, 0018, and 0019. These are validation-set results.

HOTA	DetA	AssA	MOTA	IDF1	Precision	Recall
77.48	74.31	81.04	85.52	92.37	97.32	88.02

Output

output/
├── tracking_kitti/
└── rosbag/

The output directories are created automatically.

Citation

If this project is useful in your research, please cite the related work:

@article{feng2023fgomot,
       title   = {Accurate and Real-Time 3D-LiDAR Multi-Object Tracking Using Factor Graph Optimization},
       author  = {Feng, S. and Li, X. and Yan, Z. and others},
       journal = {IEEE Sensors Journal},
       year    = {2023}
}

@article{feng2025lvimot,
       title   = {LVIMOT: Accurate and robust LiDAR-visual-inertial localization and multi-object tracking in dynamic environments via tightly coupled integration},
       author  = {Feng, S. and Li, X. and Yan, Z. and others},
       journal = {ISPRS Journal of Photogrammetry and Remote Sensing},
       year    = {2025},
       volume  = {230},
       pages   = {675--692}
}

@article{li2024liolot,
       title   = {LIO-LOT: Tightly-Coupled Multi-object Tracking and LiDAR-Inertial Odometry},
       author  = {Li, X. and Yan, Z. and Feng, S. and others},
       journal = {IEEE Transactions on Intelligent Transportation Systems},
       year    = {2024}
}

@article{feng2023vimot,
       title   = {VIMOT: A Tightly-Coupled Estimator for Stereo Visual-Inertial Navigation and Multi-Object Tracking},
       author  = {Feng, S. and Li, X. and Xia, C. and others},
       journal = {IEEE Transactions on Instrumentation and Measurement},
       year    = {2023}
}

@article{feng2024LVMOT,
       title   = {Tightly Coupled Integration of LiDAR and Vision for 3D Multiobject Tracking},
       author  = {Feng, S. and Li, X. and Yan, Z. and others},
       journal = {IEEE Transactions on Intelligent Vehicles},
       year    = {2024}
}