nazzzz5265/drone_tracking

---
license: mit
task_categories:
- object-detection
tags:
- drone
- detection
- yolo
- kalman-filter
pretty_name: UAV Drone Detection Dataset
---

# UAV Drone Detection Dataset

Drone bounding box detections extracted from UAV video footage using a fine-tuned YOLOv8n model.

## Dataset Details

- **Format:** Parquet
- **Rows:** 1,406 detections
- **Source videos:** 2 drone videos (828 and 2580 frames)
- **Model:** YOLOv8n fine-tuned on [pathikg/drone-detection-dataset](https://huggingface.co/datasets/pathikg/drone-detection-dataset) (3,000 images, 30 epochs)
- **Confidence threshold:** 0.1

## Columns

| Column | Type | Description |
|---|---|---|
| `video_name` | str | Source video identifier |
| `frame_number` | int | Frame index in the video |
| `image` | bytes | Raw detection frame image |
| `num_detections` | int | Number of drones detected in frame |
| `bbox_x1/y1/x2/y2` | int | Bounding box coordinates (pixels) |
| `center_x/y` | float | Bounding box center coordinates |
| `confidence` | float | Model detection confidence (0–1) |
| `class_name` | str | Detected class (`drone`) |

## Pipeline

1. Videos downloaded from YouTube using `yt-dlp`
2. Frames extracted at 5 fps using `ffmpeg`
3. Drone detection using fine-tuned YOLOv8n
4. Tracking using Kalman filter