ISSN 2394-3777 (Print)

ISSN 2394-3785 (Online)

Available online at www.ijartet.com

International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)

Vol. 7, Issue 4, April 2020

Fire Detection and Alert System using

Convolutional Neural Network
Mathivanan R1, Prem kumar M2, Sakthivel B3, Vadivel M4
Student, Information Technology, Francis Xavier Engineering College, Tirunelveli, India 1
Student, Information Technology, Francis Xavier Engineering College, Tirunelveli, India 2
Student, Information Technology, Francis Xavier Engineering College, Tirunelveli, India 3
Student, Information Technology, Francis Xavier Engineering College, Tirunelveli, India 4

Abstract: Fire is one of the leading hazards endangering human life and property around the world, may lead to a
considerable material damage and serious injury and even death. In order to avoid such losses, we widely practiced to use
smoke and flame detectors. However, realization of video surveillance system is more reasonable, because it allows
monitoring large areas and open space that the traditional systems of fire and smoke control cannot do. In this project fire is
detected using surveillance camera and the alert message along with captured fire image is sent to the Owner / Incharge of
the property through an email and also the system produces an alarm in order to take earliest actions and thereby avoid high
property loss.

Keywords: Camera, Detection, Image processing, Recognition

developed. In this dissertation, the image processing
I. INTRODUCTION technology is used to replace the traditional detector to
Fire monitoring and protection has always been an collect, analyze and process the image of fire scene of large
external hot spot, especially some large occasions and space, and finally achieves the purpose of real-time fire
important areas, such as the warehouse, forest, substation, detection and recognition. Fire monitoring is mainly divided
railway tunnel etc. Once these places fires, the consequences into Fire detection and flame detection. In the process of
will be disastrous. The existing fire detection technology is combustion, the fire not only has distinctive color
mainly carried out in the process of sensitive detection of characteristics, but also has very important morphological
fire, gas and temperature, these sensors, such as Fire sensors, characteristics. Fire is one of the leading hazards
have the advantages of high sensitivity, fast response, strong endangering human life and property around the world. To
ant interference ability, low cost, long service life and wide avoid large scale fire and Fire damage, some point-type
application. But in the open-space environment, because of thermal and Fire detectors are widely used, but such
high, large space, air mobility and other reasons, Fire, gas detectors need to be in a close proximity of the fire and
and temperature in the process of transmission of these easily to be fail or damaged in a bad environment. Along
signals easily fade away, so the fire signal which finally with the progress of computer vision and image processing,
arrived the detector is very weak, making the Fire, video-based fire detection is currently a fairly common
temperature, gas and other detectors to reduce the detection technology which has remarkable advantages over
accuracy, easy to delay the best time to fire alarm, fire traditional methods, such as fast response and wide detection
disaster hazards to enlarge. Especially some open places like area. Fire is the forecasting symbol of fire, so Fire detection
forests, use the Fire sensor to detector fire is simply provides earlier fire range than flame detection. The
impossible. Therefore, for large-space environment, there is traditional Fire detection or classification methods can be
a need to monitor the fire in other ways. In order to summarized as two steps, First, calculating manual features
overcome the shortcomings of the traditional fire detection, which may be the color, texture, shapes, irregularity, flutter,
with the development of computer vision, digital image or frequency from the input Fire images Second, training a
processing and pattern recognition technology, video-based classifier based on the extracted features to test whether an
fire detection technology has been gradually studied and image is Fire. In this way, the performance of these methods

All Rights Reserved © 2020 IJARTET 14

 ISSN 2394-3777 (Print)
ISSN 2394-3785 (Online)
Available online at www.ijartet.com

International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)

Vol. 7, Issue 4, April 2020

bag strategy. Chen and Huang [4] proposed a Gaussian

depend on whether the manual features are reasonable and model to simulate HSV and analyze the time and space
human often rely on experience in the choice of features, factors of the flame, but the Gaussian mixture model
which is blindness, cumbersome and complex. Although requires higher calculation time and the analysis is fuzzier.
some methods have achieved good results, these manual However, the utilization of the sensor has limitations thanks
features are designed for specific data, and if the same to the various properties of the combustion product. Burnett
features are used to deal with different data sets, the results and Wing [5] used a new low-cost camera that can reduce
may not be satisfactory. Nowadays, deep learning can learn the interference of smoke on the flame and has excellent
representations of data with multiple levels of abstraction detection capabilities for RGB and HSV, but there are still
and automatically learn. These advanced methods have some limitations within the recognition and application of
dramatically improved the state-of-the-art in visual object this camera. Töreyin et al. [6] proposed a Gaussian mixture
recognition. Motivated by the success of deep learning, a background estimation method for detecting motion pixels in
Fire detection method based on Convolution Neural video. This method selects candidate fire regions by color
Networks is proposed which can automatically learn features model then performs wavelet analysis in time and space
from the original two-dimensional images. This method can domains to figure out high frequency activity within the
solve the shortcomings of traditional methods of fire region. Almost like the previous problem, this method has
detection. high computational complexity in practical applications. Han
et al. [7] used motion detection supported the Gaussian
II. LITERATURE SURVEY model and multicolor model and obtained good experimental
For fire detection, the normal method is to use a sensor results. However, since Gaussian models and color models
for detection. One of the defects is that the high false rate require an outsized amount of computational time, they can
because the trigger alarm is predicated on the concentration not be applied to actual scenes. Chen et al. [8] improved the
of particles or the encompassing temperature and is therefore traditional flame detection method, and thus the flame
easily disturbed by the surroundings. At an equivalent time, flickering detection algorithm is incorporated into the
this method cannot know the situation and therefore the real- scheme to detect fires in color video sequences. Testing
time status of the hearth. For outdoor scenes, which are results show that the proposed algorithms are effective,
notable fire hazards, this sort of sensor cannot provide robust, and efficient. However, the calculation speed is slow,
effective detection. Due to the various problems in suitable for 320 ∗ 240 images. It’s going to not be suitable
traditional fire identification, the way to accurately for top quality.Dimitropoulos et al. and Çetin et al. [9, 10]
identify the hearth has received great attention. Therefore, used a variety of flame characteristics to judge and achieved
fire detection has achieved rapid development within good results.
the direction of fireside detection sensors, improvement in Research on video analysis for fire detection has
detection equipment, and fire detection supported video become a hot topic in computer vision. However, the
images. conventional algorithms use exclusively rule-based models
Khan et al. [1] proposed a method based on video using and features vector to classify whether a frame is fire or not.
flame dynamics and static indoor flame detection, using the These features are difficult to define and depend largely on
color, perimeter, area, and roundness of the flame. Their the kind of fire observed [1]. The outcome leads to low
method takes a little fire like a candle as an unimportant detection rate and high false-alarm rate. A different approach
part. By removing then applying the flame growth for this problem is to use a learning algorithm to extract the
characteristics to gauge, this method may have an useful features instead of using an expert to build them. In
enormous problem in early fire warning. Seebam rungsat et this paper, author proposed a convolutional neural network
al. [2] proposed a rule based on the combination of HSV and (CNN) for identifying fire in videos. Convolutional neural
YCbCr. Their system requires extra conversion of color network are shown to perform very well in the area of object
space and is therefore better than using only one color space classification. This network has the ability to perform feature
method, but their work only uses the static characteristics of extraction and classification within the same architecture.
the flame. The method is relatively fragile and not stable Tested on real video sequences, the proposed approach
enough. Foggia et al. [3] proposed a novel motion detection achieves better classification performance as some of
method based on the disordered nature of the flame-word relevant conventional video fire detection methods and

All Rights Reserved © 2020 IJARTET 15

 ISSN 2394-3777 (Print)
ISSN 2394-3785 (Online)
Available online at www.ijartet.com

International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)

Vol. 7, Issue 4, April 2020

indicates that using CNN to detect fire in videos is very be subdivided into these two analysis components and a
promising. classifier component.

III. PROPOSED SYSTEM 4) Alarm Module

A. Architecture Diagram The alarm module is concerned with raising an

alarm at the detection of fire in the frame under
consideration. This module continuously checks for fire
pixels in the final frame submitted by the classifier
component. Once the potential fire frame is detected, it
raises an alarm to denote the presence of fire.

B. Packages used
NumPy is that the fundamental package for scientific
computing in Python. It is a Python library that gives a
multidimensional array object, various derived objects (such
as masked arrays and matrices), and an assortment of
routines for fast operations on arrays, including
mathematical, logical, shape manipulation, sorting,
selecting, I/O, discrete Fourier transforms, basic algebra ,
basic statistical operations, random simulation and far more.
At the core of the NumPy package, is that the ndarray object.
This encapsulates n-dimensional arrays of homogeneous
data types, with many operations being performed in
compiled code for performance.
Pandas is an open source Python package that
Fig. 1 Architecture diagram
gives numerous tools for data analysis. The package comes
1) Extract Images Frame from video with several data structures which will be used for
This module is concerned with the video data several different data manipulation tasks. It also has a
processing required for the system to work. Its major role in variety of methods that can be invoked for data analysis,
the system is to read the video data and extract image frame which comes in handy when working on data science and
machine learning problems in Python.
from video. Matplotlib may be a Python 2D plotting library which
2) Color Conversion Module produces publication quality figures during a sort
of hardcopy formats and interactive environments across
The video may use different formats or platforms. Matplotlib can be used in Python scripts, the
configurations for the processing of raw video data. For the Python and shells, the Jupiter notebook, web application
system to work, it needs the data to be homogeneous with servers, and four graphical user interface toolkits. Matplotlib
the same format and configuration. This module converts tries to form easy things easy and hard things possible. You
the video data to RGB format, which makes further can generate plots, histograms, power spectra, bar charts,
processing of video data easier. errorcharts, scatterplots, etc., with just a couple of lines of
. 3) Fire Detector Module code.
This module is a crucial component of the system.
Convolutional neural network are shown to perform very
well in the area of object classification .It is concerned with
the pixel and frame analysis, which are two basic
methods used in the classification of fire pixels from
background pixels and non-fire pixels. Thus this module can

All Rights Reserved © 2020 IJARTET 16

 ISSN 2394-3777 (Print)
ISSN 2394-3785 (Online)
Available online at www.ijartet.com

International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)

Vol. 7, Issue 4, April 2020

IV. EXPERIMENTAL RESULTS

The main aim of this study is to automatically detect fire
in frame that was extracted from video, using computer
vision methods, implemented in real-time with the help of
the OpenCV library. Proposed solution must be applicable in
existing security systems, meaning with the utilization of
normal industrial or personal video cameras.

Fig. 3 Fire detection using camera

The above Fig. 3 shows that the hearth is detected using

Fig. 2 Code for fire detection Camera and produces alarm sound, necessary solution
precondition is that camera is static. Given the pc vision and
The above Fig. 2 shows the code for detection of fire
image processing point of view, stated problem corresponds
which uses convolutional neural network (CNN) for
to detection of dynamically changing object, supported this
identifying fire in videos.
color and moving features. While static cameras are utilized,
c. Overview of Technique background detection method provides effective
CNNs use relatively little pre-processing compared to segmentation of dynamic objects in video sequence.
other image classification algorithms. This means that the Candidate fire-like regions of segmented foreground objects
network learns the filters that in traditional algorithms are determined consistent with the rule-based color
were hand-engineered. This independence from prior detection.
knowledge and human effort in feature design is a major
advantage.
When programming a CNN, the input is a tensor with shape
(number of images) x (image width) x (image height) x
(image depth). Then after passing through a convolutional
layer, the image becomes abstracted to a feature map, with
shape (number of images) x (feature map width) x (feature
map height) x (feature map channels).

All Rights Reserved © 2020 IJARTET 17

 ISSN 2394-3777 (Print)
ISSN 2394-3785 (Online)
Available online at www.ijartet.com

International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)

Vol. 7, Issue 4, April 2020

The above Fig.6 shows the alert email message along with
captured fire image.so that the user can contact the fire
department immediately and take remedial actions as soon
as possible.

V. CONCLUSION
It is concluded that the fire is detected, alarm is produced
and the alert message with captured fire image is
immediately sent to the owner through an email.So that the
owner can immediately contact the fire department and take
actions easily and thereby avoiding high property loss. The
advantage of using this system is that the cost of using this
type of detection is cheaper and the implementation of this
type system is greatly simpler compare to those traditional
methods. Secondly the response time of detection system is
quicker compare to the opposite traditional detection
Fig. 4 Visualization of fire range using histogram methods since a vision camera- based fire detection system
doesn't required any type conditions to trigger the camera
The above Fig.5 shows the range of fire and it is used to and it has the ability to monitor a large area depends on the
analyze the range of fire easily. Along with the progress of camera used.
computer vision and image processing, video-based fire
detection is currently a fairly common technology which has REFERENCES
remarkable advantages over traditional methods, such as fast [1]. R. A. Khan, J. Uddin, S. Corraya, and J. Kim, “Machine vision based
response and wide detection area. indoor fire detection using static and dynamic features,” International
Journal of Control and Automation, vol. 11, no. 6, 2018.
[2]. [2] J. Seebamrungsat, S. Praising, and P. Riyamongkol, “Fire
detection in the buildings using image processing,” in Proceedings of
the 2014 Third ICT International Student Project Conference (ICT-
ISPC), pp. 95–98, IEEE, Bangkok, Thailand, March 2014.
[3]. P. Foggia, A. Saggese, and M. Vento, “Real-time fire detection for
video-surveillance applications using a combination of experts based
on color, shape, and motion,” IEEE Transactions on Circuits and
Systems for Video Technology, vol. 25, no. 9, pp. 1545–1556, 2015.
[4]. L. Chen and W. Huang, “Fire detection using spatial-temporal
analysis,” in Proceedings of the World Congress on Engineering, pp.
3–5, London, UK, July 2013.
[5]. J. D. Burnett and M. G. Wing, “A low-cost near-infrared digital
camera for fire detection and monitoring,” International Journal of
Remote Sensing, vol. 39, no. 3, pp. 741–753, 2018.
[6]. B. U. Töreyin, Y. Dedeoğlu, U. Güdükbay, and A. E. Çetin,
“Computer vision based method for real-time fire and flame
detection,” Pattern Recognition Letters, vol. 27, no. 1, pp. 49–58,
2006.
[7]. X. F. Han, J. S. Jin, M. J. Wang, W. Jiang, L. Gao, and L. P. Xiao,
“Video fire detection based on Gaussian mixture model and multi-
color features,” Signal, Image and Video Processing, vol. 11, no. 8,
pp. 1419–1425, 2017.
[8]. J. Chen, Y. He, and J. Wang, “Multi-feature fusion based fast video
flame detection,” Building and Environment, vol. 45, no. 5, pp. 1113–
1122, 2010.
Fig.5 Alert email message along with captured fire [9]. K. Dimitropoulos, P. Barmpoutis, and N. Grammalidis, “Spatio-
image. temporal flame modeling and dynamic Texture analysis for automatic
video-based fire detection,” IEEE Transactions on Circuits and
Systems for Video Technology, vol. 25, no. 2, pp. 339–351, 2015.

All Rights Reserved © 2020 IJARTET 18

 ISSN 2394-3777 (Print)
ISSN 2394-3785 (Online)
Available online at www.ijartet.com

International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)

Vol. 7, Issue 4, April 2020

[10]. A. E. Çetin, K. Dimitropoulos, B. Gouverneur et al., “Video fire

detection—review,” Digital Signal Processing, vol. 23, no. 6, pp.
1827–1843, 2013.

BIOGRAPHY

Mathivanan R pursuing B.Tech (IT)

degree in Francis Xavier Engineering
College, Tirunelveli, TamilNadu, India. He
has participated in various hackathons and
project contests conducted by various
organizations. He is awarded INSPIRE
(Innovation in Science Pursuit for Inspired
Research) award for the year 2013-2014.
He won various prizes in coding contests.
He has attended various guesttalk and
Conferences in various colleges. He is
awarded as Best Outgoing Student,
Department of Information Technology for
the year 2016-2020

Prem kumar M pursuing B.Tech (IT)

degree in Francis Xavier Engineering
College, Tirunelveli, TamilNadu, India.
He has participated in various hackathons
and project contests conducted by various
organizations. He had co-ordinated various
events for junior students.

Sakthivel B pursuing B.Tech (IT) degree

in Francis Xavier Engineering College,
Tirunelveli ,TamilNadu, India.

Vadivel M pursuing B.Tech (IT) degree in

Francis Xavier Engineering College,
Tirunelveli, TamilNadu, India.

All Rights Reserved © 2020 IJARTET 19