UAV Object Detection with YOLOv8

This repository contains the first version of an object detection system developed for a UAV (Unmanned Aerial Vehicle) project during an internship. The system is built using YOLOv8 and a custom-trained model trained on a dataset from Roboflow Universe. It detects a variety of objects, including cars, cell phones, water bodies, fire, birds, planes, rockets, buses, persons, laptops, and more. Designed for real-time detection, it has been tested with a webcam and is intended for UAV applications such as environmental monitoring, surveillance, or navigation.

Features

• Real-time object detection using YOLOv8.
• Custom-trained model (best.pt) for diverse object classes.
• Webcam input for testing, adaptable for UAV camera feeds.
• Lightweight and extensible for drone deployment.

Directory Structure

uav-object-detection/
├── Yolov8m/
│   └── weights/
│       └── best.pt     # Custom-trained YOLOv8 model
├── detect.py           # Main script for real-time detection
├── args.yaml           # Configuration file for YOLOv8 training/inference
├── Yolov8m.pt          # Pretrained YOLOv8 medium model
├── README.md           # Project documentation
└── requirements.txt    # Python dependencies

Prerequisites

• Python 3.8+
• OpenCV (cv2)
• Ultralytics YOLOv8
• A webcam or video file for testing
• (Optional) NVIDIA GPU with CUDA for faster inference

Installation

1. Clone the repository:

   git clone https://github.com/BinaryFighter-01/uav-object-detection.git
   cd uav-object-detection

2. Set up a virtual environment (recommended):

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

3. Install dependencies:

   pip install -r requirements.txt

   If requirements.txt is missing, install core libraries:

   pip install ultralytics opencv-python

4. Verify model file: Ensure Yolov8m/weights/best.pt is present in the repository. This is the custom-trained model.

Usage

1. Run the detection script:

   python detect.py

   This will:

   • Load the Yolov8m/weights/best.pt model.
   • Start real-time detection using your webcam (default input).
   • Display the video feed with detected objects annotated.

2. Customize input: To use a video file instead of a webcam, modify the cv2.VideoCapture line in detect.py:

   cap = cv2.VideoCapture('path/to/your/video.mp4')

3. Exit: Press q while the video window is active to stop detection.

File Descriptions

• detect.py: Main script for running real-time object detection.
• Yolov8m/weights/best.pt: Custom-trained YOLOv8 model for your dataset.
• Yolov8m.pt: Pretrained YOLOv8 medium model (used as a starting point for training).
• args.yaml: Configuration file for training or inference settings (e.g., model parameters, dataset paths).
• requirements.txt: Lists Python dependencies required for the project.

Model Details

• Base Model: YOLOv8 (medium variant, fine-tuned).
• Training Data: Custom dataset from Roboflow Universe.
• Classes: Cars, cell phones, water bodies, fire, birds, planes, rockets, buses, persons, laptops, and more.
• Weights: best.pt (located in Yolov8m/weights/).

Notes for UAV Deployment

• The current script is designed for testing on a local machine. For UAV integration:
  • Replace webcam input with a drone camera feed (e.g., RTSP stream: rtsp://<drone-ip>/stream).
  • Optimize for lightweight models (e.g., YOLOv8n) if deploying on edge devices like Jetson Nano.
  • Account for UAV hardware constraints (power, compute).
• Use args.yaml to adjust inference settings (e.g., confidence threshold) for better performance.

Future Improvements

• Add support for UAV-specific video streams (e.g., RTSP, MAVLink).
• Implement object tracking for consistent detection across frames.
• Optimize model for edge devices (e.g., quantization, pruning).
• Include model evaluation metrics (e.g., mAP, precision, recall).

Contributing

This is an internship project, but contributions are welcome! To contribute:

1. Fork the repository.
2. Create a feature branch (git checkout -b feature-name).
3. Submit a pull request with a clear description.