FireSafe: Real-Time Fire and Person Detection with Automated Notification Alerting Using CCTV Surveillance Footage

Overview

FireSafe is a real-time mobile-first safety solution that detects fire and persons using live CCTV feeds. It leverages the lightweight YOLOv11n deep learning model trained on the D-Fire and Human datasets to provide efficient, accurate detections. When a fire or person is detected, users are instantly notified via mobile alerts and audible device beeps. The end-to-end pipeline consists of:

• A Flutter-based mobile app for viewing live feeds and alerts
• A Flask backend that processes CCTV RTSP streams in real-time
• Firebase Cloud Messaging (FCM) for sending notifications

📱 Demo

▶️ Click here to watch the Mobile App Demo

Features

• Live multi-camera feed monitoring
• Real-time fire and person detection
• Alert notifications via Firebase
• Secure user authentication and feed management
• Annotated live stream using MJPEG rendering

Directory Structure

FireSafe/
├── backend/                # Flask backend for inference and streaming
│   ├── app.py              # Main app with routes for inference
│   ├── background.py       # Background thread for stream processing
│   ├── model/              # Trained YOLOv11n weights and inference utils
│   └── requirements.txt    # Python dependencies
├── frontend/               # Flutter mobile app
│   ├── lib/                # Dart source code
│   ├── pubspec.yaml        # Flutter dependencies
├── README.md               # Project documentation
└── demo/                   # Sample videos, screenshots, and results

Installation

Frontend

1. Install Flutter SDK.
2. Connect a device or emulator.
3. Navigate to the frontend folder and run:

cd frontend/
flutter clean
flutter pub get
flutter run

Backend

1. Ensure Python 3.8+ is installed.
2. Create a virtual environment (optional but recommended).
3. Install backend dependencies and start the server:

cd backend/
pip install -r requirements.txt
python app.py
python background.py

Note: Inference runs on YOLOv11n using the Tapo TP-Link C212 camera RTSP stream.

How to Run

1. Launch the backend server first (app.py and background.py).
2. Start the frontend Flutter app on your mobile device.
3. Log in, add camera feeds, and view live detections in real-time.
4. Notifications are automatically pushed when fire/person is detected with confidence above the set threshold.

Model Details

Model	[email protected]	[email protected]:0.95	Latency
YOLOv8n	0.743	0.426	61 ms
RTMDet-tiny	0.754	0.435	57 ms
YOLOv11n	0.768	0.446	54 ms

YOLOv11n outperformed all baselines in both detection accuracy and latency, making it ideal for real-time mobile deployment.

Datasets Used

• D-Fire Dataset — Annotated real-world CCTV fire images
• Human Dataset — Person detection under diverse indoor and outdoor scenes

Complete Demo

You can Watch our complete system in action: Demo Video on Google Drive