Computer Vision-Based Fire

Detection System

Department of Artificial Intelligence

Erick 艾瑞克
Mitro Satu( 萨图 )
Minidu ( 杜满 )
Awais Khan ( 吴华 )
 Outline
• Introduction
• Fire Detection System
• Motivation
• Literature Survey
• Proposed Model
• Dataset
• Experimental Study and Results
• Conclusion and Future works
• References

2
 Introduction
• Fire is the most perilous and hazardous abnormal event with the
potential for significant human, structural, and economic losses.
• Even a small fire can have a severe impact on temperature-sensitive
industries like Chemical, Oil, and Gas.
• Fire detection is crucial for public safety and property damage
prevention.
• Recent progress in computer vision and deep learning enables the
development of accurate detection systems.
• The proposed fire detection model, trained on a specialized dataset,
effectively identifies fire incidents.

3
Fire Detection System

4
 Motivation
 Fire disasters are significant threats to human lives and
property.
 Limitations in traditional sensor-based methods.
 Sensor deployment is impractical in certain scenarios.
 Challenges arise when attempting to adapt to vast and
complex environments, particularly in large, open spaces.
 Other challenges include
 variations in lighting conditions,
 color overlap between fire and similar objects
 the complexity of detecting fire at its initial stage, especially on
extremely sunny days

6
 Literature Survey
Authors Method
Frizzi et al. Convolutional Neural Network
Praveen et al. Convolutional Neural Network
Khan et al. Fine-tuning of Convolutional Neural
Network
Salim et al. Inception-V2 and Faster R-CNN
Nori et al. YOLO (You Only Look Once)
Suhas et al. Simplified YOLO version
Proposed Model

8
16
Proposed CNN
Architecture
 Proposed CNN
Architecture
• Architecture includes convolutional layers for
feature extraction, capturing corners and edges.
• Rectified Linear Unit (ReLU) activation nullifies
negative inputs in the convolutional layer.
• Pooling layers optimize efficiency, reducing
spatial volume after convolution.
• Max pooling operates independently between
convolution layers on each feature map.
• Fully connected layers and softmax/logistic layer
for image classification and final output.
Dataset
 Experimental Setup
• Software
– Python framework
– Keras and Tensorflow deep learning library
• Hardware
– Windows 11 operating system utilizing a i5 11th gen
– NVIDIA Graphics RTX1650 GPU
– minimum of 8GB RAM

19
 Experimental Objectives
• To assess the performance of the proposed CNN
model for classifying malware using
Accuracy
Loss
Precision
Recall
F1-score
• To visualize the classification performance using
Accuracy/Loss graph

21
Experimental Results
Datas No. of Accurac Los Precisio Recall F1-
et Epoch y s n score
s (%)

Fire 50 95.67 0.19 0.96 0.96 0.96

Adam Optimizer with a learning rate of

22
Accuracy/Loss Graphs

Higher accuracy
Lower loss
Does not suffer overfitting

23
Precision-Recall Curve

23
ROC CURVE

23
Fire Detection by the
Proposed System
Outputs for Non-Fire
Detection
 Conclusion and Future
Work
• Fire are detected from images and live webcam using
the proposed CNN.
• The proposed CNN produced promising results with
an accuracy of 97% for 100 epochs.
• As future work, precise real-time recognition can be
concentrated
• Also, Challenges in model accuracy arise due to the
scarcity of fire datasets, with only less number of
images available for training
 References
[1] Sebastien Frizzi,Rabeb Kaabi, Moez Bouchouicha,Jean-Marc Ginoux, Eric Moreau
and Farhat Fnaiech. ”convolutional neural network for video fire and smoke
detection”. IEEE conference, pp.877-882, 2016.
[2] Praveen Sankarasubramanian and Dr. Ganesh.E.N. ”CNN based Intelligent
Framework to Predict and Detect Fire”. vol.20:pp.755–772, 2022.
[3] Khan muhammad , jamil ahmad,irfan mehmood , seungmin rho andsung wook
baik. ”Convolutional Neural Networks Based Fire Detection in Surveillance Videos”.
IEEE conference, vol.6:pp.18174–18183, 2018.
[4] Senthilnayaki, B., Venkatalakshami, K., Dharanyadevi, P., Nivetha, G., & Devi, A.
(2022, March). An efficient medical image encryption using magic square and PSO.
In 2022 International Conference on Smart Technologies and Systems for Next
Generation Computing (ICSTSN) (pp. 1-5). IEEE
[5] Salim Said AL-Ghadani and Jayakumari.c. ”Innovating Fire Detection System Fire
using Artificial Intelligence by Image Processing”. International Journal of
Innovative Technology and Exploring Engineering (IJITEE) ISSN: 2278-3075 (Online),
vol.9:pp.349–356, 2020.
[6] Ruba Nori.R, Rabah N. Farhan and Safaa Hussein Abed. ”Indoor and outdoor fire
localization using YOLO algorithm”. Journal of Physics:Conference Series, pp.1-10,
2021.
21
 References
[7] Xiaojun Bai and Zongxin Wang. ”A research on forest fire detection technology based on
deep learning”. International Conference on Computer Network, Electronic and Automation
(ICCNEA), pp.85-90, 2021.
[8] Senthilnayaki B, Rajeswary C, Nivetha G, Dharanyadevi P, Mahalakshmi G and Devi A ,
"Traffic Sign Prediction and Classification Using Image Processing Techniques," 2022
International Conference on Smart Technologies and Systems for Next Generation Computing
(ICSTSN), Villupuram, India, 2022, pp. 1-5, doi: 10.1109/ICSTSN53084.2022.9761286.
[9] Dongqing Shen, Xin Chen, Minh Nguyen and Wei Qi Yan. ”Flame detection using deep
learning”. 4th International Conference on Control, Automation and Robotics, pp.416-420,
2018.
[10] Senthilnayaki, B., Keerthana, S., & Sowmya, H. (2021). Automatic License Plate
Detection And Recognition. International Advanced Research Journal in Science, Engineering
and Technology, 8(8), 22-26.
[11] Turgay Celik, Huseyin Ozkaramanlı and Hasan Demirel. ”Fire and smoke detection
without sensors”. 15th European Signal Processing Conference (EUSIPCO), pp.1794-1798,
2007.
[12] Suhas G , Chetan Kumar, Abhishek B S, Digvijay Gowda K A and Prajwal R. ”Fire
Detection Using Deep Learning”. International Journal of Progressive Research in Science and
Engineering, Vol.1:pp.1–5, 2020.

28
Thank
You