Real-Time-Smoke-Fire-Detection-YOLO11

📌 Table of Contents

1. Introduction
2. Video Demonstration
3. Dataset
4. Training Summary
5. Example Detections
6. Installation and Usage
   • Installation
   • Inference

Safeguard Lives with Smart Fire and Smoke Detection – A New Era of Protection 🛡️

Problem Description

Fire incidents can cause catastrophic damage and loss of life when not detected early enough. Current detection methods, which rely primarily on traditional sensors, often have delayed response times that can lead to severe consequences. This creates a critical need for faster, more reliable detection systems that can identify fire hazards at their earliest stages. To address this challenge, this project implements advanced deep learning technology to detect fires and smoke as they emerge, enabling rapid response and intervention.

Solution

Flare Guard represents a state-of-the-art approach to fire safety monitoring. At its core, the system uses YOLOv11, a powerful object detection model, to continuously analyze video feeds for signs of fire and smoke in real-time. When potential threats are detected, the system immediately sends alerts through Telegram/WhatsApp messaging platforms, ensuring that stakeholders are notified without delay.

Why Choose Flare Guard?

Key capabilities:

• Real-time video processing with 30-60 frames per second
• High-accuracy threat identification (80.6% Precision)
• Robust performance across various environmental conditions
• Immediate alert system with visual confirmation via Telegram and WhatsApp.
• Adaptable for multiple use cases including surveillance systems, industrial monitoring, home security, and robotic applications

The combination of speed, accuracy, and reliable notification makes Flare Guard particularly effective for environments where early fire detection is critical for safety and asset protection.

Video Demonstration

Dataset

The dataset consists of 10,463 annotated images, available on Roboflow. This dataset is designed for training and evaluating object detection models tailored for real-time fire and smoke detection. It is suitable for:

• Surveillance systems (CCTV monitoring, smart security cameras)
• Industrial safety applications (factories, warehouses, refineries)
• Residential safety solutions (smart home fire detection)
• Autonomous monitoring systems (drones, robotics, IoT devices)

Split	Images	Annotations
Training	9,156	27,468
Validation	872	2,616
Test	435	1,305

Classes: Fire, Smoke
Annotation Format: YOLOv11-compatible bounding boxes

# Download dataset via Roboflow
from roboflow import Roboflow
rf = Roboflow(api_key="YOUR_API_KEY")
project = rf.workspace("sayed-gamall").project("fire-smoke-detection-yolov11")
dataset = project.version(2).download("yolov11")

Training Summary

The model was trained using YOLOv11 on a dataset of fire and smoke images. Training stopped early due to no improvement over 20 epochs, with the best results observed at Epoch 92.

Training Graph

Final Validation Results

Metric	Value
Precision (P)	0.806
Recall ®	0.717
mAP@50	0.770
mAP@50-95	0.492

Class-Specific Performance

Class	Precision	Recall	mAP@50	mAP@50-95
Fire	0.813	0.806	0.828	0.513
Smoke	0.800	0.629	0.711	0.472

Example Detections

Here are examples from the test set:

Installation and Usage 🚀

Installation

1. Clone the repository:

   git clone https://github.com/sayedgamal99/Real-Time-Smoke-Fire-Detection-YOLO11
   cd Real-Time-Smoke-Fire-Detection-YOLO11
   

2. Install the required packages:

   pip install ultralytics
   

Inference

To perform inference with the trained model on test images, run:

yolo detect predict model=models/best_nano_111.pt source=data/house.png conf=0.35 iou=0.1

and there’s the output:

To perform inference in real-time using a webcam:

yolo detect predict model=models/best_nano_111.pt source=0 conf=0.35 iou=0.1 show=True

---