TRIBHUVAN UNIVERSITY

Institute of Engineering
Pulchowk Campus, Pulchowk

A
THESIS PROPOSAL
ON

Fire Detection Using Image Processing Techniques

Submitted By:
Rabin Kasula (068/MSI/f/615)
Symbol No. – 12651

Submitted To:
Department of Electronics and Computer Engineering
11th December 2017
 THESIS PROPOSAL

ON

Fire Detection Using Image Processing Techniques

Rabin Kasula (068/MSI/f/615)

Symbol No. – 12651

Department of Electronics and Computer Engineering

IOE, Pulchowk Campus, Pulchowk

11th December 2017

 Acknowledgement

I would like to express my sincere gratitude to Mr. Roshan Koju, (Phd Student in IoE) for
his encouragement and precious guidance during the thesis title selection. I am thankful
to our program coordinator Dr. Basanta Joshi, our Head of Department Dr. Dibakar Raj
Panta and Mr. Baburam Dawadi for providing suitable platform to prepare this proposal.
I would also like to thank my seniors, friends and faculty of Department of Electronics
and Computer Engineering for providing me their views and ideas regarding thesis work.

i
TABLE OF CONTENTS

ACKNOWLEDGEMENT ....................................................................................................i
TABLE OF CONTENTS .................................................................................................... ii
LIST OF FIGURES ............................................................................................................ iii
LIST OF EQUATIONS ...................................................................................................... iv
LIST OF ABBREVIATION ............................................................................................... v
1. INTRODUCTION ........................................................................................................... 1
1.1. BACKGROUND AND MOTIVATION ................................................................................ 1
1.2. PROBLEM STATEMENT .................................................................................................. 2
1.3. OBJECTIVES ................................................................................................................... 2
1.4. SCOPE OF WORK ............................................................................................................ 3
2. LITERATURE REVIEW ................................................................................................. 4
3. METHODOLOGY ........................................................................................................... 6
3.1. Procedures...................................................................................................................... 6
3.2. FLOWCHART .......................................................................................................... ……9
3.3. TOOLS............................................................................................................................11
3.4. IMAGE COLLECTION....................................................................................................11
4. EXPECTED OUTPUT ....................................................................................................12
5. WORK SCHEDULE .......................................................................................................13
6. REFERENCES ................................................................................................................14

ii
List of Figures

Figure 1: to YCbCr Conversion Matrix for analog image................................7

Figure 2: RGB to YCbCr Conversion Matrix for digital image.......................7
Figure 3: Flowchart of Fire Detection…………………………………………...10
Figure 4: Thesis Time Schedule ………………………………………………...13

iii
List of Equations

Equation 1: Rule I ...................................................................................... 7

Equation 2: Rule II...................................................................................... 8
Equation 3: Mean Calculation of Y, Cb and Cr.......................................... 8

iv
List of Abbreviation and Symbols

CCTV Closed Circuit Television

UAV Unmanned Ariel Vehicle
ROI Region of Interest
SLIC Simple Linear Iterative Clustering
2D 2 Dimensional

v
1. Introduction
1.1. Background and Motivation

Fire is the rapid oxidation of material in the exothermic chemical process of combustion,
releasing heat, light, and various reaction products. We use fire very often in cooking
food, warming rooms, and other different works. It is most commonly used in our daily
lives. The most common hazard due to fire in our daily lives is the house fire and forest
fire. Fire causes great harm to the human beings and also responsible for the destruction
of property and a very serious economic losses.

With the growing population and increasing households, the risks due to fire have
increased day by day. Application of fire detection as a tool has been increased due to the
frequent occurrence of extended fire with consequences on human health and security.
For minimization and alertness, implementation of fire detection systems using different
smoke and heat sensors are common nowadays. The accuracy, reliability and positional
distributions of the sensor determine the betterment of the system. For high precision fire
detection large numbers of sensors are needed in the case of outdoor applications.
Sensors also need a frequent battery charge which is impossible in a large open space.
Sensors detect fire if and only if it is close to fire. This will lead to damaging of sensor
[1].

In order to mitigate such problems, nowadays digital image processing tools are being
used for fire detection. Nowadays, Closed Circuit Televisions (CCTVs) are common and
cheap tool used for capturing the images and which can be connected further to the
Computer based image processing systems in order to detect the fire nearby. Further, the
output of the image processing systems can be linked to the alarming devices to send fire
alert message to nearby people and also can be linked to fire controlling devices to
control the fire. Some of the big companies also use the Unmanned Arial Vehicles
(UAVs) to capture the images of their companies’ premises.

These type of systems offer several distinguish advantages over those traditional
detection methods. For example the cost of using this type of detection is cheaper and the
implementation of this type of system is greatly simpler compare to those traditional
methods. Secondly the response time of fire detection system is faster compare to any
other traditional detection methods since a vision sensor-based fire detection system does
not required any type conditions to trigger the sensors and it has the ability to monitor a
large area depends on the camera used. The most benefit of this type of system is the fire
source can be saved in a form of image or video which can be used for promoting the
diversification of the fire detection method greatly [2].

1
In computer based digital image processing systems, the captured image is needed to
convert into the form compatible to the machine or tool. Then, based on the techniques
used the colored or RGB image is converted to HIS/HSV or CMYK/CMY or YCbCr or
YUV or binary color models. Further, the converted images are processed to detect the
region with fire based on the color threshold technique, texture feature extraction
technique etc. Also, the edge detection techniques or algorithms are applied to the
original images to detect the boundaries or edges of the objects present in the image. By
combining and applying the segmentation techniques, the image with only fire will be
detected. The final output image can be used to calculate the fire intensity and fire
volume in the captured area. Based on this fire image the alarming systems and fire
controlling systems can be activated.

1.2. Problem Statement

Fire detection using image processing is better than using the traditional approach of
using sensors. Though it is better technique, there exist many more complexities in
analyzing the images containing the fire like objects. For those cases the texture
classification used in [3] need to be implemented along with the color model for accurate
detection of fire which brings the complexity in generation of fire detection algorithms.

Also, the images taken from the CCTV Cameras will not be always in good form, it may
contain noise (blurred) which is also a problem for detecting actual fire image in the
received image. The solution is to use the median filters to the original images as used in
[4].

With the use of better algorithms and techniques for image analysis, color classifications,
edge detections and segmentation step the output accuracy and the efficiency can be
improved. For example: Canny edge detection technique has better outputs than Sobel
edge detection technique and instead of ROI based segmentation used in [2]; Watershed
algorithm can be used for better output.

1.3. Objectives

 To use image processing techniques for detecting fire in household and city
environments and locations based on images captured.
 To use image processing techniques for detecting fire in forest environments and
locations based on images captured.

2
1.4. Scope of Work

Currently, fire outbreak is common issue happening in Nepal and the damage caused by
these types of incidents is tremendous toward nature and human interest. Use of image
based fire detection system using CCTV cameras and image processing systems has
significant uses in Houses, Apartments, City areas, Forests, Shopping Malls etc.

The satellite images can be used to analyze the areas under fire and send alerts to the
people residing nearby. Furthermore, the CCTV cameras in large cities can also be
integrated to fire detection systems so as to evacuate the city before the fire expands
tremendously.

Since the traditional approach of fire detection using sensors is somehow costly and has
delayed response, this image based fire detection technique significantly reduce the cost
and have quick response.

3
2. Literature Review
There are several researches that have been carried out in this area. Some of the
researcher uses the live videos for image capturing and some uses the sampled images.

Poobalan and Liew developed the algorithm in [2] that converted the RGB images to the
HSI color model. Further, the color threshold technique and Sobel edge detection
followed by ROI based segmentation was used for extracting the fire contained portion
from the whole image.

In case of noisy images, an improved and best approach of using adaptive median filter
for removing the noises in the image and the advanced color recognization algorithm
based on the fire color that changes from white to yellow, orange and red, color threshold
and edge thinning was used in [5] for detecting the fire in the image.

The individual extracted values of R, G and B were used to detect fire color in the image.
An YCbCr color space was used for histogram plot and based on the histogram
segmentation was done by Tawade and Patane in [6]. It also considered the motion and
flame flicker of the fire.

Premal and Vinsley [1] used the YCbCr color model for separating luminance from
chrominance and also calculated the mean and standard deviation of Y, Cb and Cr for
creating their own four rules. Two rules were used for segmenting fire region and two
were used for segmenting high temperature region which is a best approach to find the
strength of the fire.

A frequency analysis approach and color model based segmentation technique was
implemented by Gomes, Santana and Barata in [7] for fire detection. It also used the
context-based gating of the background learning processes for reducing the chances of
erroneously learning moving objects.

In order to detect forest fire, Yuan, Liu, and Zhang in [4] developed the algorithm that
used the median filter for noise suppression, used Lab Color model for fire segmentation,
used Otsu threshold technique for image segmentation and erosion & dilation technique
of morphology for better outputs. They also integrated the blob counter for fire tracking.

A new concept of fire detection by mixing color classification and texture classification
was introduced by Chino, Avalhais, Rodrigues and Traina in [3], that discriminates the
actual fire from the fire like objects. The research used the Naive-Bayes classifier for
color classification and KNN classifier for texture classification and finally the two
results are mixed to get final output of fire image. It also included the SLIC (Simple
4
Linear Iterative Clustering) superpixel algorithm to generate superpixels and LBP (Local
Binary Pattern) for texture extractions.

Zaidi, Lokman, Daud, Achmad and Chia in [8], compared the accuracy of fire detection
or recognization using RGB and YCbCr Color models with various images taken at
morning, night, day and evening. The researchers also considered the image filtering
technique for removing noises from original images. Seven sets of rules (2 rules for RGB
and 5 rules for YCbCr) were developed for identifying the fire in the image and finally a
feature extraction and fire pixel classification were performed for assuring the fire.

5
3. Methodology

3.1 Procedures
Fire detection using image processing techniques include following steps:
a. Image Acquisition
b. Application of Median Filter
c. Use of Color Model and Conversion
d. Edge Detection
e. Combining Two Results
f. Image Segmentation to Get Final Result

a. Image acquisition

This step involves image capturing via digital camera and transferring into computers to
receive the image in its digital format. Image format used will be in JPEG image
compression standard and RGB color model. The data will be collected in a day light
which is morning, afternoon and evening. More data will be collected for a better
performance. A fire will be set up in different scenarios so that various data could be
collected.

b. Application of Median Filter

Median filter is a nonlinear digital filtering technique, often used to remove noise from an
image or signal. Such noise reduction is a typical pre-processing step to improve the
results of later processing (for example, edge detection of an image). Median filtering is
very widely used in digital image processing because, under certain conditions, it
preserves edges while removing noise. The main idea of the median filter is to run
through the signal entry by entry, replacing each entry with the median of neighboring
entries. The pattern of neighbors is called the "window", which slides, entry by entry,
over the entire signal. For 1D signal, the most obvious window is just the first few
preceding and following entries, whereas for 2D (or higher-dimensional) signals such as
images, more complex window patterns are possible (such as "box" or "cross" patterns).
Further, median filter is based upon moving a window over an image and computing the
output pixel as the median of the gray values within the input window. If window is J x K
in size then we can order the J*K pixels from smallest to largest in the gray level values.
If J*K is odd then the median will be:

(J*K+1)/2

And this will be the entry in the list of ordered gray values.

6
 c. Use of Color Model and Conversion

The image obtained after Median filter will be RGB color. For attaining better result and
reliable recognizing the region of interest in image processing, the RGB color model will
be converted to YCbCr color space or model. The formula for RGB to YCbCr conversion
is:

Figure 1: RGB to YCbCr Conversion Matrix for analog image

And for JPEG images or digital images the formula is:

Figure 2: RGB to YCbCr Conversion Matrix for digital image

where Y, Cb, Cr = 0 to 255 (8 bits per sample).

After conversion, two rules will be applied to detect fire section in captured image. The
rules will be:
I. For fire region we can say that R>G>B in RGB color space or model whereas
this can be translated in to YCbCr color space as Y>Cb i.e. the intensity of Y
component is greater than the intensity of Cb component. Then it can be
explained as follows:

( , ), ( , )> ( , )
( , )=
0, ℎ
Equation 1: Rule I

Where I(x,y) represents the input RGB image. Y(x,y) and Cb(x,y) are
luminance and chrominance Blue values at different special locations (x,y).
Ri(x,y) is the pixel which satisfies Rule I.

7
 II. Since flame region is the brightest region in scene and chrominance red (Cr)
component in the fire region is more the mean values of Y and Cr channels in
the overall image (Ymean and Cr mean) contains valuable information. In fire
region, the value of Y component at each special location is greater than mean
value of the Y component (Ymean) and the value of Cr component at each
special location is greater than mean value of the Cr component (Cr mean).
However, for the pixels in the non fire region one of these two conditions are
not satisfied. The rule can be expressed mathematically as:

( , ), (& , ) > , >

( , )=
0, ℎ

Equation 2: Rule II

Where Rii(x,y) is the pixel in the input image which satisfies both Rule I and
Rule II. Mean values can be calculated as:

1
Ymean = Y(x, y)
M x N

1
Cbmean = Cb(x, y)
M x N

1
Cr mean = Cr(x, y)
M x N

Equation 3: Mean Calculation of Y, Cb and Cr

where M x N is the total number of pixels in the input image.

d. Edge Detection

Edge detection includes a variety of mathematical methods that aim at identifying points
in a digital image at which the image brightness changes sharply or, more formally, has
discontinuities. The points at which image brightness changes sharply are typically
organized into a set of curved line segments termed edges. Edge detection is a
fundamental tool in image processing, machine vision and computer vision, particularly
in the areas of feature detection and feature extraction. The growth of the fire within the
images will be detected and the edges will be separated using Canny Edge Detection. The
process of Canny edge detection algorithm can be broken down to 5 different steps:

8
 1. Apply Gaussian filter to smooth the image in order to remove the noise
2. Find the intensity gradients of the image
3. Apply non-maximum suppression to get rid of spurious response to edge
detection
4. Apply double threshold to determine potential edges
5. Track edge by hysteresis: Finalize the detection of edges by suppressing all the
other edges that are weak and not connected to strong edges.

e. Combining Two Results

By combining the results obtained from color model conversion with rules and edge
detection, the clearly separated fire image will be obtained.

f. Image Segmentation to Get Final Result

Image segmentation is used to separate the fire region of the image from non fire region
in the image. Watershed model for segmentation will be used to separate fire from the
combined image for obtaining the final result.

3.2 Flowchart
The flowchart of the overall methodology can be seen as:

9
Figure:3 Flowchart of Fire Detection

10
3.3 Tools

 Matlab R2017a
 Digital Camera

3.4 Image Collection

Capturing the images with fire using digital camera and collect the fire images in
different scenarios from internet.

11
4. Expected Output
At the end of this research, a new algorithm for fire detection based on image processing
techniques will be generated that can detect the fire from the images captured by
cameras. The accuracy and efficiency of the fire detection will be improved compared to
others.

12
5. Work Schedule

Weeks
Description of Works 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Literature Survey
Developing Algorithm
Implement Algorithm
Prepare Test Conditions
Output Analysis
Documentation

Figure:4 Thesis Time Schedule

13
6. References
[1] C. Emmy Prema1, S. S. Vinsley, “Image Processing Based Forest Fire Detection
Using YCbCr Colour Model”, International Conference on Circuit, Power and
Computing Technologies [ICCPCT], ©2014 IEEE.

[2] Kumarguru Poobalan and Siau-Chuin Liew, “Fire Detection Algorithm Using
Image Processing Techniques”, Proceeding of the 3rd International Conference
on Artificial Intelligence and Computer Science (AICS2015), 12 - 13 October
2015, Penang, Malaysia.

[3] Daniel Y. T. Chino, Letricia P. S. Avalhais, Jose F. Rodrigues Jr., Agma J. M.

Traina, “BoWFire: Detection of Fire in Still Images by Integrating Pixel Color
and Texture Analysis”, Proceedings of the 28th SIBGRAPI Conference on
Graphics, Patterns and Images, 2015, ©2015 IEEE.

[4] Chi Yuan, Zhixiang Liu, and Youmin Zhang, “UAV-based Forest Fire Detection
and Tracking Using Image Processing Techniques”, International Conference on
Unmanned Aircraft Systems (ICUAS), Denver Marriott Tech Center Denver,
Colorado, USA, ©2015 IEEE.

[5] Patil A.D., Kuwar K.R., More M.S., Pagar P.A. & Somawanshi S.D., “Advance
Algorithm for Fire Detection Using Image Processing and Color Recognition”,
IOSR Journal of Electronics and Communication Engineering (IOSR-JECE) e-
ISSN: 2278-2834, p-ISSN: 2278-8735 PP 09-14.

[6] Hemangi Tawade , R.D. Patane, “Opmized Fire Detection Using Image
Processing Based Techniques”, International Journal of Science, Engineering and
Technology Research (IJSETR), Volume 4, Issue 6, June 2015.

[7] Pedro Gomes, Pedro Santana and Jose Barata, “A Vision-based Approach to Fire
Detection”, International Journal of Advanced Robotic Systems, 2014.

[8] Norsyahirah Izzati binti Zaidi, Nor Anis Aneza binti Lokman, Mohd Razali bin
Daud, Hendriyawan Achmad and Khor Ai Chia, “Fire Recognization Using RGB
and YCbCr Color Space”, Asian Research Publishing Network (ARPN) Journal of
Engineering and Applied Sciences, VOL. 10, NO. 21, November 2015.

14