README.md

---
license: apache-2.0
task_categories:
- object-detection
tags:
- computer-vision
- object-tracking
- multi-camera
- uav
---

# Multi-Camera Multi-Vehicle Tracking System (UAV Dataset)

This dataset contains synchronized multi-UAV video footage and comprehensive tracking annotations, accompanying the paper **[A Topology-Aware Spatiotemporal Handover Framework for Continuous Multi-UAV Tracking](https://arxiv.org/abs/2605.15779)**.

It is designed for evaluating real-time multi-camera multi-vehicle tracking (MCMT) systems, focusing on solving trajectory fragmentation and maintaining global identity persistence across isolated UAV fields of view.

- **Project Page:** [https://www.jye.me/ICUAS2026/](https://www.jye.me/ICUAS2026/)
- **GitHub Repository:** [https://github.com/JYe9/multi-camera-multi-vehicle-tracking-system](https://github.com/JYe9/multi-camera-multi-vehicle-tracking-system)

<p align="center">
  <img src="https://huggingface.co/datasets/jye9/Multi-Camera-Multi-Vehicle-Tracking-System/resolve/main/assets/Pipeline.png" alt="Multi-Camera Multi-Vehicle Tracking Pipeline" width="80%"/>
</p>

## Dataset Contents

### 1. Video Assets
Three synchronized 1-minute video clips captured from different UAV perspectives in complex urban environments.
- `uav1_1min.mp4`
- `uav2_1min.mp4`
- `uav3_1min.mp4`

<p align="center">
  <img src="https://huggingface.co/datasets/jye9/Multi-Camera-Multi-Vehicle-Tracking-System/resolve/main/assets/uav1.jpg" alt="UAV1" width="32%"/>
  <img src="https://huggingface.co/datasets/jye9/Multi-Camera-Multi-Vehicle-Tracking-System/resolve/main/assets/uav2.jpg" alt="UAV2" width="32%"/>
  <img src="https://huggingface.co/datasets/jye9/Multi-Camera-Multi-Vehicle-Tracking-System/resolve/main/assets/uav3.jpg" alt="UAV3" width="32%"/>
<br/>
<em>Example: Consistent global tracking (e.g., the red car with ID 70) across overlapping UAV fields of view.</em>
</p>

### 2. Annotations (CSVs)
The dataset includes structured CSV files with rich metadata generated by our tracking pipeline. 

**Base Schema:**
```text
frame_index, x_min, y_min, x_max, y_max, class_id, confidence, tracker_id, global_id, camera_id, speed, status
```

*Note: Speed is estimated in km/h via pixel-to-meter scaling. Status tracks active/new_entry/exiting states.*

## Sample Usage

After cloning the [GitHub repository](https://github.com/JYe9/multi-camera-multi-vehicle-tracking-system) and installing the requirements, you can run the pipeline with the following command:

```bash
python main.py \
  --source_video_path_uav1 videos/uav1_1min.mp4 \
  --target_video_path_uav1 output_videos/1mins_UAV1_output.mp4 \
  --target_csv_path_uav1 csv/1mins_UAV1.csv \
  --source_video_path_uav2 videos/uav2_1min.mp4 \
  --target_video_path_uav2 output_videos/1mins_UAV2_output.mp4 \
  --target_csv_path_uav2 csv/1mins_UAV2.csv \
  --source_video_path_uav3 videos/uav3_1min.mp4 \
  --target_video_path_uav3 output_videos/1mins_UAV3_output.mp4 \
  --target_csv_path_uav3 csv/1mins_UAV3.csv \
  --model_path models_zoo/3UAVs.pt \
  --confidence_threshold 0.5 \
  --iou_threshold 0.5
```

## Citation

If you find this dataset or our framework useful in your research, please cite our paper:

```bibtex
@misc{ye2026topologyaware,
      title={A Topology-Aware Spatiotemporal Handover Framework for Continuous Multi-UAV Tracking}, 
      author={Jianlin Ye and Christos Kyrkou and Panayiotis Kolios},
      year={2026},
      eprint={2605.15779},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}
```