Helmet Detection System

A real-time helmet detection web app powered by a custom-trained YOLOv8 model. Upload an image or video and the system draws bounding boxes around detected helmets, persons, and motorcycles — instantly telling you whether a helmet is present or not.

---

Demo

Helmet Detected ✅	No Helmet ❌
Green bounding box drawn around helmet	Red alert — no helmet found in frame

---

✨ Features

• 🖼️ Image detection — upload JPG, PNG, or WEBP and get an annotated result instantly
• 🎥 Video detection — upload MP4/MOV and every frame is processed; annotated video returned
• 🟢 Clear verdict — "Helmet detected" or "No helmet detected" shown prominently
• 📦 Bounding boxes — color-coded per class (helmet, person, motorcycle)
• 📊 Detection table — confidence scores and box coordinates for every detection
• ⬇️ Download result — save the annotated image or video directly from the UI

---

🧠 Model

• Architecture: YOLOv8n (nano — fast inference)
• Dataset: Open Images V7 (via FiftyOne)
• Classes: helmet · person · motorcycle
• Training: 35 epochs · 640px · batch 32 · Google Colab GPU
• Weights file: best.pt (not included in repo — see setup below)

---

🗂️ Project Structure

helmet-detection/
├── backend/
│   ├── main.py              # FastAPI app — detection endpoints
│   ├── requirements.txt     # Python dependencies
│   └── weights/
│       └── best.pt          # ← place your model weights here
└── frontend/
    ├── src/
    │   ├── App.tsx
    │   ├── components/      # DropZone, ResultPanel, DetectionTable …
    │   ├── hooks/           # useDetection
    │   └── utils/           # API calls, TypeScript types
    ├── package.json
    └── vite.config.ts

---

🚀 Getting Started

Prerequisites

• Python 3.10+
• Node.js 18+
• Your trained best.pt weights file

---

🔌 API Endpoints

Method	Endpoint	Description
GET	/health	Health check + loaded classes
POST	/detect/image	Upload image → annotated image + detections
POST	/detect/video	Upload video → annotated video + frame stats

🛠️ Tech Stack

Layer	Technology
Model	YOLOv8n (Ultralytics)
Backend	FastAPI + Uvicorn
Image processing	OpenCV
Frontend	React 18 + TypeScript
Bundler	Vite
Icons	Lucide React

---

📓 Training Notebook

The model was trained in Google Colab using:

• FiftyOne to download and prepare Open Images V7
• YOLOv8n fine-tuned for 35 epochs on helmet/person/motorcycle classes
• Export to YOLO format → trained with ultralytics trainer

See notebook/final.ipynb for the full training pipeline.

---

🤝 Contributing

Pull requests are welcome. For major changes, open an issue first to discuss what you'd like to change.

---

📄 License

MIT License — feel free to use, modify, and distribute.