Fire Guard AI: Real-Time Fire & Smoke Detection System

About the Project

Fire is one of the world’s most devastating natural and industrial hazards, causing both loss of life and property. Traditional smoke detectors only react when smoke physically reaches the sensor, losing critical minutes. In large open areas (forests, warehouses, factories) and high-ceiling facilities, these sensors are often completely ineffective.

Fire Guard AI is an intelligent early warning system that uses AI-powered computer vision technology to detect fire and smoke formation in the very first second, while it can still be controlled. It connects directly to existing security cameras (IP/CCTV), requiring no additional hardware investment.

The system detects the moment a flame starts as a small spark or when light smoke begins to spread in the environment within milliseconds. Early-stage fires that traditional sensors cannot detect for minutes are visually identified by AI, and alarms are generated instantly.

Comprehensive Features and Modules

• YOLOv8-Based Real-Time Detection: The system uses a YOLOv8 object detection model specially trained on fire and smoke datasets. It detects both fire and smoke classes simultaneously with over 96% accuracy. It also performs well under night vision and low-light conditions.
• Multi-Camera Support: Dozens of RTSP camera streams can be monitored simultaneously from a single server. Independent detection thresholds, sensitivity levels, and alarm rules can be defined for each camera. Suitable for factories, warehouses, forest watchtowers, and facility management.
• Instant Multi-Channel Notification System: When fire or smoke is detected, a cropped image, confidence score, camera name, and timestamp are instantly sent to the relevant personnel’s Telegram app within seconds. Simultaneously, detailed alarm reports can be sent via email.
• Event History and Evidence Management: Each detected event is recorded in the database (SQLite/PostgreSQL) along with the video clip, detection timestamp, camera information, and confidence score. Retrospective analysis, statistical extraction, and digital evidence management are possible.
• Live Monitoring and Analytics Dashboard: Features a modern, dark mode-supported web interface built with Next.js. Offers live streaming of all cameras in grid or single view, detection history filtering, daily/weekly/monthly alarm statistics visualization, and system health monitoring.
• Zone-Based Sensitivity: Specific regions (ROI - Region of Interest) can be defined on the camera image. For example, sensitivity can be set to maximum for a factory’s chemical storage area, while minimizing false positives in controlled flame source areas like kitchens.

How It Differs from Competitors

Fire Guard AI is designed with a complete Edge Computing architecture, running directly on local servers at the facility:

1. Zero Cloud Cost: Video streams are not sent to cloud servers for analysis. No internet bandwidth consumption, no cloud processing costs.
2. Ultra-Low Latency: Thanks to local processing, the detection-to-alarm time is under 500 milliseconds. The 3-10 second delay in cloud-based solutions doesn’t exist here.
3. Full Data Privacy: Camera footage never leaves the facility. Compliant with GDPR and local data protection regulations.
4. Hardware Independent: Integrates with any camera brand supporting the RTSP protocol. No vendor lock-in.

Use Cases

• Forest Fire Early Warning: Detects forest fires while they are still in a small area by connecting to cameras on watchtowers.
• Industrial Facilities and Factories: Monitors fire risks 24/7 in chemical storage areas, petrochemical plants, and production lines.
• Smart Building and Facility Management: Transforms existing cameras in apartments, malls, and office buildings into fire safety systems.
• Warehouse and Logistics Centers: Prevents millions in inventory loss through early detection in large storage facilities.
• Agriculture and Greenhouses: Continuously monitors fire risk in agricultural areas with dense dry vegetation.

Architecture and Technical Infrastructure

The entire system is built on an asynchronous and modular architecture:

• AI & Backend Engine: Developed using Python and FastAPI. The YOLOv8 model works integrated with the Ultralytics library and OpenCV. Accelerated with TensorRT on NVIDIA GPUs, providing real-time (30+ FPS) analysis. Camera streams are processed in parallel using multiprocessing structures.
• Real-Time Communication: Detected events and live video frames are delivered to the frontend via low-latency WebSockets.
• Frontend / Management Panel: Built with Next.js 14, React, TailwindCSS, and Zustand for a high-performance and responsive interface.
• Notification Services: Telegram Bot API and SMTP email integration operate as independent asynchronous services.
• Environment Support: Can be easily deployed in Windows, Linux, or Docker environments.

GitHub Repository: github.com/vahapogut/AI-fire-detection