Civilian detection with drone vision and deep learning (YOLO, Roboflow, PyTorch).
This project is a machine learning–based computer vision system that detects civilians from aerial imagery captured by drones. It is designed to work in day and night conditions, and can handle both low and high proximity detections.
| Day/Night Detection | High/Low Proximity | |
|---|---|---|
 |
 |
- Civilian detection from aerial drone vision.
- Supports both daytime and nighttime scenarios.
- Trained with Roboflow’s Look Down Folks dataset:
👉 Roboflow Dataset - Powered by YOLOv5 + PyTorch for real-time detection.
- Exported in Jupyter Notebook (.ipynb) format for easy execution.
Clone this repository or download the .ipynb file to run the project:
git clone https://github.com/parkqdev/aerial-civilian-detection.git
While the system performs well in varied environments, there are limitations such as:
- Potential false positives (e.g., detecting objects shaped like humans).
- Reduced accuracy in very low light, unless paired with night-vision technology.
- Environmental conditions (fog, rain, heavy shadows) can impact detection reliability.
This project is developed for educational and research purposes only.
- The dataset used for training and testing was sourced from publicly available news coverage.
- The original raw footage is not included in this repository to respect copyright and ethical considerations.
The intent of this project is to explore computer vision techniques for civilian detection in aerial/surveillance contexts, not to make political commentary or commercial products.
This project is licensed under the Apache License 2.0.
See the LICENSE file for details.