Counter-Drone Detection and Tracking System

A deep learning-based system for detecting and tracking drones in video feeds using YOLOv8 Nano and PyTorch, optimized for Apple Silicon (M4 chip) with MPS support.

📋 Overview

This project implements a real-time drone detection and tracking system using:

• YOLOv8 Nano - Lightweight object detection model
• PyTorch with MPS - Optimized for Apple Silicon
• Norfair Tracker - Multi-object tracking (optional)
• OpenCV - Video processing and visualization

🚀 Features

• ✅ Real-time drone detection in video/webcam feeds
• ✅ Multi-object tracking with unique IDs
• ✅ Optimized for Apple Silicon (MPS backend)
• ✅ Configurable confidence thresholds
• ✅ Support for video files and webcam input
• ✅ Clean, modular codebase with detailed comments

📁 Project Structure

Counter Drone Detection & Tracking/
├── train_model.py          # Training script
├── detect_video.py         # Detection script (video/webcam)
├── track_video.py          # Tracking script with Norfair
├── organize_dataset.py     # Dataset organization script
├── test_camera.py          # Camera test utility
├── create_test_video.py    # Create test video from dataset
├── data.yaml               # Dataset configuration
├── requirements.txt        # Python dependencies
├── README.md              # This file
├── Drone Dataset/
│   └── drone_dataset_yolo/
│       ├── images/
│       │   ├── train/
│       │   └── valid/
│       └── labels/
│           ├── train/
│           └── valid/
└── runs/
    └── train/
        └── counter_drone/
            └── weights/
                ├── best.pt
                └── last.pt

🛠️ Installation

1. Prerequisites

• Python 3.10 or higher
• macOS with Apple Silicon (M1/M2/M3/M4)
• VS Code (recommended)

2. Create Virtual Environment

# Navigate to project directory
cd "Counter Drone Detection & Tracking"

# Create virtual environment
python3 -m venv venv

# Activate virtual environment
source venv/bin/activate

# IMPORTANT: Make sure you see (venv) in your prompt
# If you see (base) or other conda environments, deactivate them first:
# conda deactivate  # If using conda

3. Install Dependencies

# Make sure virtual environment is activated (you should see (venv) in prompt)
# If not, activate it: source venv/bin/activate

# Upgrade pip
pip install --upgrade pip

# Install PyTorch with MPS support (for Apple Silicon)
pip install torch torchvision

# Install other dependencies (this will install NumPy < 2.0 for compatibility)
pip install -r requirements.txt

# Verify NumPy version (should be < 2.0)
python -c "import numpy; print('NumPy version:', numpy.__version__)"

4. Verify Installation

# Check PyTorch MPS availability
python3 -c "import torch; print('MPS available:', torch.backends.mps.is_available())"

📊 Dataset Setup

Dataset Structure

Ensure your dataset follows this structure:

Drone Dataset/drone_dataset_yolo/
├── images/
│   ├── train/          # Training images
│   └── valid/          # Validation images
└── labels/
    ├── train/          # Training labels (YOLO format)
    └── valid/          # Validation labels (YOLO format)

Organize Dataset (If Needed)

If your dataset is in a single folder (e.g., dataset_txt/), use the organization script to split it into train/valid:

# Organize dataset into train/valid splits (80/20)
python organize_dataset.py

# Custom options
python organize_dataset.py --source "Drone Dataset/drone_dataset_yolo/dataset_txt" --target "Drone Dataset/drone_dataset_yolo" --train-ratio 0.8

This script will:

• Split images and labels into train/valid sets (80/20 by default)
• Create the required folder structure
• Copy files to appropriate directories
• Maintain image-label pairs

YOLO Label Format

Each image should have a corresponding .txt file with the same name. Label format:

class_id center_x center_y width height

All coordinates are normalized (0-1). Example:

0 0.5 0.5 0.2 0.3

Configure data.yaml

Update data.yaml if your dataset path is different:

path: ./Drone Dataset/drone_dataset_yolo
train: images/train
val: images/valid
nc: 1  # Number of classes
names:
  0: drone

🏃 Training

Train the Model

# Activate virtual environment (if not already active)
source venv/bin/activate

# Run training
python train_model.py

Training Configuration

The training script (train_model.py) is configured with:

• Model: YOLOv8 Nano (yolov8n.pt)
• Epochs: 40
• Image Size: 640x640
• Batch Size: 16 (adjust based on available memory)
• Device: Automatically detects MPS/CPU

Training Output

After training, the best model will be saved at:

runs/train/counter_drone/weights/best.pt

Training results (plots, metrics) will be available in:

runs/train/counter_drone/

🎥 Detection

Detect in Video File

# Detect drones in a video file
python detect_video.py --source test.mp4 --conf 0.25

# Use custom model
python detect_video.py --model runs/train/counter_drone/weights/best.pt --source test.mp4

Detect from Webcam

# Use default webcam (index 0)
python detect_video.py --source 0

# Use specific webcam
python detect_video.py --source 1

Command-Line Options

• --model: Path to model weights (default: runs/train/counter_drone/weights/best.pt)
• --source: Video file path or webcam index (default: 0)
• --conf: Confidence threshold 0-1 (default: 0.25)

Controls

• Press q to quit
• Press s to save current frame

🎯 Tracking

Track Drones in Video

# Track drones in a video file
python track_video.py --source test.mp4 --conf 0.25

# Track from webcam
python track_video.py --source 0

Tracking Features

• Unique IDs: Each detected drone gets a unique ID
• Trails: Visual trails show drone movement paths
• Color Coding: Different colors for different tracks
• Bounding Boxes: Detection boxes with confidence scores

Install Norfair (Optional)

If you want to use the advanced tracking features:

pip install norfair

If Norfair is not installed, the script will use a simple tracking fallback.

📝 Usage Examples

Example 1: Complete Training and Detection Pipeline

# 1. Organize dataset (if not already organized)
python organize_dataset.py

# 2. Train the model
python train_model.py

# 3. Test on video file
python detect_video.py --source test.mp4

# 4. Test with tracking
python track_video.py --source test.mp4

Example 2: Real-time Webcam Detection

# Detect drones from webcam
python detect_video.py --source 0 --conf 0.3

# Track drones from webcam
python track_video.py --source 0 --conf 0.3

Example 3: Create Test Video from Dataset

# Create a test video from validation images
python create_test_video.py

# Custom options
python create_test_video.py --output my_test_video.mp4 --fps 15 --duration 10

# Then use it for detection
python detect_video.py --source my_test_video.mp4

Example 4: Custom Confidence Threshold

# Higher confidence (fewer false positives)
python detect_video.py --source test.mp4 --conf 0.5

# Lower confidence (more detections)
python detect_video.py --source test.mp4 --conf 0.15

Example 5: Test Camera Access

# List all accessible cameras
python test_camera.py --list

# Test specific camera
python test_camera.py --camera 0

🔧 Troubleshooting

Issue: MPS not available

Solution: Make sure you're using PyTorch 2.0+ and macOS 12.3+ with Apple Silicon.

pip install --upgrade torch torchvision

Issue: Dataset not found

Solution: Check that data.yaml points to the correct dataset path and the folder structure matches the expected format.

Issue: Out of memory during training

Solution: Reduce batch size in train_model.py:

BATCH_SIZE = 8  # or lower

Issue: Video file not opening

Solution:

• Make sure the video file path is correct
• Use absolute path instead of relative path
• Verify file exists: ls -la path/to/video.mp4
• Check file format is supported (MP4, AVI, MOV, MKV)
• Try a different video file to rule out corruption
• Example: python detect_video.py --source /absolute/path/to/video.mp4

Issue: NumPy 2.x Compatibility Error

Error Message:

A module that was compiled using NumPy 1.x cannot be run in NumPy 2.3.3
AttributeError: _ARRAY_API not found

Solution:

This happens when using conda base environment with NumPy 2.x. Fix it:

1. Use Virtual Environment (NOT conda base):

   # Deactivate conda if active
   conda deactivate
   
   # Activate venv (you should see (venv) not (base))
   source venv/bin/activate

2. Fix NumPy Version:

   # Option 1: Use the fix script
   ./fix_numpy_issue.sh
   
   # Option 2: Manual fix
   pip uninstall numpy opencv-python opencv-contrib-python -y
   pip install -r requirements.txt

3. Verify:

   python -c "import numpy; print('NumPy:', numpy.__version__)"  # Should be 1.x.x
   python -c "import cv2; print('OpenCV:', cv2.__version__)"     # Should work

Issue: Webcam not working

Solution:

• macOS Camera Permissions:

  1. Go to System Settings > Privacy & Security > Camera
  2. Enable camera access for:
     • Terminal (if running from terminal)
     • Python (if available in the list)
     • VS Code (if running from VS Code)
  3. Restart Terminal/VS Code after enabling permissions
  4. Test camera access: python test_camera.py --list

• Test Camera Access:

  # List all accessible cameras
  python test_camera.py --list
  
  # Test specific camera
  python test_camera.py --camera 0

• Other Solutions:

  • Try different webcam indices (0, 1, 2, etc.)
  • Make sure no other application is using the webcam
  • Use a video file instead: python detect_video.py --source video.mp4

📈 Performance Tips

1. Training: Use MPS backend for faster training on Apple Silicon
2. Inference: Reduce image size for faster processing
3. Tracking: Adjust distance_threshold in track_video.py for better tracking
4. Confidence: Tune confidence threshold based on your use case

📚 Additional Resources

• YOLOv8 Documentation
• PyTorch MPS Documentation
• Norfair Tracker
• OpenCV Documentation

🤝 Contributing

Feel free to submit issues, fork the repository, and create pull requests for any improvements.

📄 License

This project is open source and available for educational and research purposes.

👤 Author

Built for Counter-Drone Detection and Tracking System

🙏 Acknowledgments

• Ultralytics for YOLOv8
• PyTorch team for MPS support
• Norfair for tracking capabilities

---

Happy Drone Detecting! 🚁