Edge-Deployable UAV Target Tracking System

An offline, air-gapped computer vision pipeline that simulates drone-based target tracking on low-power edge hardware.

This project trains a YOLOv8-nano model on the VisDrone aerial dataset, mathematically optimizes the weights via ONNX, and runs real-time inference strictly on CPU hardware without PyTorch or GPU dependencies.

🚀 Features

• Edge-Optimized Inference: Uses onnxruntime (CPUExecutionProvider) to run complex vision models on standard laptop or embedded CPUs.
• Air-Gapped Execution: Operates entirely offline. No cloud APIs or internet connection required for inference.
• Defense-Grade Pipeline: Simulates the hardware constraints of onboard drone flight computers.
• High-Accuracy Classification: Detects 10 specific aerial target classes including vehicles, pedestrians, and bicycles using the VisDrone standard.

🛠️ Architecture

This project is divided into two distinct environments:

1. Cloud Training (Google Colab / GPU): * Fetches the VisDrone dataset via Ultralytics.
   • Trains yolov8n.pt using an NVIDIA T4 GPU.
   • Exports the trained graph to best.onnx to strip heavy framework overhead.
2. Edge Simulation (Local Windows / CPU):
   • Ingests the ONNX model and raw video feeds.
   • Preprocesses frames using pure OpenCV and NumPy.
   • Runs inference, applies Non-Maximum Suppression (NMS), and outputs annotated tracking coordinates in real-time.

📦 Prerequisites

You do not need a GPU or PyTorch to run the simulation. You only need:

• Python 3.8+
• A test video of drone footage named test_video.mp4

⚙️ Installation & Setup

1. Clone the repository:

   git clone [https://github.com/YOUR_USERNAME/UAV-Edge-Tracker.git](https://github.com/YOUR_USERNAME/UAV-Edge-Tracker.git)
   cd UAV-Edge-Tracker

2. Create and activate a virtual environment:

   python -m venv venv
   # On Windows:
   venv\Scripts\activate
   # On Linux/Mac:
   source venv/bin/activate

3. Install dependencies:

pip install -r requirements.txt

Usage

1. Ensure inference.py, best.onnx, and test_video.mp4 are in the same directory.
2. Execute the tracking pipeline:

python inference.py

(Press q to safely terminate the video stream).

Technologies Used

• Model: YOLOv8-nano
• Dataset: VisDrone
• Graph Optimization: ONNX
• Inference Engine: ONNX Runtime
• Computer Vision: OpenCV (cv2)
• Matrix Math: NumPy

License

This project is licensed under the MIT License.