Abstract Science. All rights reserved. 1. Introduction Fire

Abstract
Fire incidence is one of the major disasters of human society. This paper proposes a still imagebased
fire detection system. It has many advantages like lower cost, faster response, and large coverage.
The existing methods are not able to detect fire region adequately. The proposed method overcome and
addresses the issue. A binary contour image of flame that is capable of classifying fire or no fire in image
for fire detection is proposed in this study. The color of fire area can range from red yellow to almost white.
So, here it is challenges the detected area is actually fire or no fire. Our propose method consists of five
parts. Firstly, the digital image is taken from dataset and the digital image is sampled and mapped as a
grid of dots or picture elements. We convert image to separate RGB Color range Matrix. We define some
rules to select yellow color range of the image later on converted the image to binary range. Finally, binary
contour image of flame information that detect the fire. We have analyzed different types of fire images in
different varieties and found accuracy 85-90%.
Keywords: dataset, digital image, binary range and matrix, binary contour image, fire detection

Copyright © 2016 Institute of Advanced Engineering and Science. All rights reserved.
1. Introduction
Fire is one of the biggest disasters for human begins. It is very challenging for detecting
fire, environmental disasters or serious damage to human life. In particular, accidents involving
fire and explosion have attracted interest to the development of automatic fire detection
systems. Existing solutions are based on ultraviolet and infrared sensors, and usually explore
the chemical properties of fire and smoke in particle samplings 1. However, the main constraint
of these solutions is that sensors must be set near to the fire source, which brings complexity
and cost of installation and maintenance, especially in large open areas.
Several methods regarding to fire detection on videos have been proposed in the last
years. These methods use two steps to detect fire. First, they explore the visual features
extracted from the video frames (images); second, they take advantage of the motion and other
temporal features of the videos 2. In the first step, the general approach is to create a
mathematical/rule-based model, defining a sub-space on the color space that represents all the
fire-colored pixels in the image.
2. Literature Review
There are several empirical models using different color spaces as RGB 1, YCbCr 3,
CIE Lab 4 and HSV 5. In these cases, the limitation is the lack of correspondence of these
models to fire properties beyond color. The problem is that high illumination value or reddishyellowish
objects lead to a higher false-positive rate. These false-positives are usually
eliminated on the second step through temporal analysis. In contrast to such methods, our
proposal is to detect fire in still images, without any further (temporal) information, using only
visual features extracted from the images. To overcome the problems aforementioned, we
propose a new method to detect fire in still images that is based on the combination of two
approaches: pixel-color classification and texture classification. The use of color is a traditional
approach to the problem; whilst, the use of texture is promising, because fire traces present
IJEECS ISSN: 2502-4752 ?
Fire Detection in Still Image Using Color Model (Hira Lal Gope)
619
particular textures that permit to distinguish between actual fire and fire-like regions. We show
that, even with just the information present in the images, it is possible to achieve a high
accuracy level in such detection.
Fire detectors are one of those amazing inventions that, because of mass production,
cost practically nothing. Recently, several methods have been proposed, with the aim to
analyze the videos acquired by traditional video surveillance cameras and detect fires or smoke,
and the current scientific effort 6, 7 focused on improving the robustness and performance of
the proposed approaches, so as to make possible a commercial exploitation. Although a strict
classification of the methods is not simple, two main classes can be distinguished, depending
on the analyzed features: color based and motion based. The methods using the first kind of
features are based on the consideration that a flame, under the assumption that it is generated
by common combustibles, such as wood, plastic, paper, or others, can be reliably characterized
by its color, so that the evaluation of the color components in RGB (Red, Green, Blue), YUV
(Luminance, Chrominance) or any other color space is adequately robust to identify the
presence of flames. This simple idea inspires several recent methods: for instance, in 8 and
9, fire pixels are recognized by an advanced background subtraction technique and a
statistical RGB color model: a set of images have been used and a region of the color space
has been experimentally identified. So that if a pixel belongs to this particular region, then it can
be classified as fire.
Generally, current residential fire detection research focuses on upholstered
furniture/mattress fires. The fire losses from residential furniture fires may decrease due to the
development of new regulations; therefore it is imperative to evaluate the new detection
approaches with the next most significant fire losses in residential fires.
The existing method cannot detect fire region properly; however, many other features
have to be taken into consideration. In our research, our propose method that can overcome
these issues. A novel feature extraction method that is capable of classifying an object as fire or
no fire in video frame for fire detection is propose in this study. The color of fire area can range
from red yellow to almost white. So, here it is challenges the detected fire is actually fire or not.
Irregularity of the boundary of the fire-colored region is taken into consideration and image is
converted to gray scale image. Eventually, our approach can identify more relevant concepts for
detecting fire by utilizing system especially, the techniques of convert images to binary images.
In our paper we have worked with a sample image in Figure 1.
Figure 1. Sample Input Image to Detect Fire
Our proposed method detects not only the fire but also it can detect the intensity of fire
like low fire, medium fire and no fire. When the flame is getting more violent, flame change their
shapes more rapidly. Therefore, the variation of flame is needed to measure the intensity of
flames. Here, contour information of the binary image is needed.
Since, the contour information of the flame is needed, the corresponding binary image,
b(x, y), of a contour image, c(x, y), is defined as follows:
( ) {
( )

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

? ISSN: 2502-4752
IJEECS Vol. 3, No. 3, September 2016 : 618 – 625
620
In the contour image, set the remaining fire color is white and it is called the binary
contour image. Now, the difference ( ), of two binary contour image,
( ) and
( ) is
as follows:
( ) |
( )
( )|
The image ( ) is called the difference contour image. After obtaining a contour
difference image, it is time to measure the intensity of fire by using the amount of white pixels.
White pixels ratio, can be defined as follows:

Where, is the amount of white pixels and n is the total number of pixels in the contour
image ( ).
The ratio is higher it means the more intensity of fires. Here, we consider three types of
fires such as small fire, medium fire and big fire.
If the ratio is equal to 0 it means no fire is detected in the image.

Second condition confirms that small fire is recognized. Similarly, third condition tells
that the medium fire is detected. And finally, fourth condition is assumed that the big fire is
detected. In our research, we exploit the hierarchical structure and their relations together with
binary images order to identify and predict more specific concept for fire detection.
This test image shows the area of contour fire pixels.
Figure 2. Detected Fire Area
In our experimental result we see that the accuracy rate is nearly 90%
3. Proposed Method
This section covers the detail of the previously proposed fire detection methods. It is
assumed that the image capturing device produces its output in RGB format. During an
occurrence of fire, smoke and flame can be seen. With the increasing in fire intensity, smoke
and flame will be visible. An easy way to comply with the conference paper formatting
requirements is to use this document as a template and simply type your text into it. So, in order
to detect the occurrence of fire, both flame and smoke need to be analyzed. Many researchers
used unusual properties of fire such as color, motion, edge, shape. Lai, et al., 10 suggested
IJEECS ISSN: 2502-4752 ?
Fire Detection in Still Image Using Color Model (Hira Lal Gope)
621
that features of fire event can be utilized for fire detection in early stages. Han, et al., 11 used
color and motion features while Kandil, et al., 12 and Liu, et al., 13 utilized shape and color
features to detect an occurrence of fire.
The main aspects of this research are to develop system with good accuracy. The
research is ongoing, and some proposals are under consideration as complements to the
currently planned approach. Using fire detection basic algorithm we are:
a) At first find out digital image from image dataset
b) After finding digital images then convert these images to RGB (Red, Green, Blue)
color range matrix
c) Select yellow color range of photo from RGB color range matrix and convert image
to binary range
d) After converting to binary image then count how much set pixel inside the image
e) Take the decision from set count value, fire is present or not.
The proposed fire detection method can be divided into five major parts: (1) Collected
Image from video frame.(2) Convert the image to RGB color, (3) Selection of yellow color range
of the image, (4) convert image to binary range, and (5) Detect the intensity of fire by contour
binary image, as depicted in Figure 3.
Figure 3. Flow Chart of Proposed Algorithm for Fire Detection in Image Sequences
Table 1. Summary For Fire and No Fire Images
Image Name Height Width Test bit Ratio Comments
fire4444 669 1000 75007 0.1121 High Range
fire14 1200 1600 282 0.00014 Actually no fire
fire3 2896 1944 342869 0.0609 High Range
fire 335 423 4209 0.0297 Medium Range
fire10 533 517 154321 0.56 High Range
fire444 211 239 0 0 No fire
fire4 282 425 32091 0.2678 High Range
fire6 2592 3872 22006 0.0022 Low Range
fire805 300 400 11860 0.0988 Medium Range
fire7 823 1291 0 0 No fire
4. Experemental Result and Discussion
We performed experiments using a dataset of fire images. It consists of different images
with various resolutions. Also, it was divided in two categories: some images containing fire, and
some images without fire. The fire images consist of emergency situations with different fire
incidents, as buildings on fire, industrial fire, car accidents, and riots. These images were
manually cropped by human experts. The remaining images consist of emergency situations
with no visible fire and also images with fire-like regions, such as sunsets, and red or yellow
objects.
? ISSN: 2502-4752
IJEECS Vol. 3, No. 3, September 2016 : 618 – 625
622
We have taken two RGB image frames then algorithm is applied on it, and result is
shown as in Figure 4(a), Figure 4(b) and Figure 4(c). Sample RGB image frames having fire, it
contains sub images of different steps in algorithm: 1st image frame, 2nd image frame having
flame, red component of fire pixel according to condition as mentioned above, motion is
detected between these two frames, and last sub image shows the fire pixel detected in image.
(a) First Image Shows the Intermediate Result of Processing, and Second Image Shows the
Contour Fire Pixels
(b) First Image Shows the Intermediate Result Of Processing, and Second Image Shows the
Contour Fire Pixels
Low Range

Medium Range

High Range
(c) These Images Show the Result of Different Ranges
Figure 4. Analytical View for Different Images
IJEECS ISSN: 2502-4752 ?
Fire Detection in Still Image Using Color Model (Hira Lal Gope)
623
5. Results
In our research there are 2 classes of image. At first step we have applied folding
method on two classes of images that classes are Fire and No Fire class and we see that our
proposed method based fire detection gives a good result. Each of the classes contains 10
images with different verities. The result is in Table 2.
Table 2. Number of Class Accuracy
Class Name Number of Image Fire Detected Not detected
Fire 10 8 2
No Fire 10 3 7
Table 3. Results per Class Success
No. of image Classes Name No. of images Fire Detected Fire Not Detected Accuracy
10 Fire 10 8 2 80.00%
10 No Fire 10 1 9 90.0%
Figure 5. Success and Failure per Class Figure 6. Accuracy Graph of Different Class
Cross-validation, sometimes called rotation estimation, is a technique for assessing how
the results of a statistical analysis will generalize to an independent data set. It is mainly used in
settings where the goal is prediction, and one wants to estimate how accurately a predictive
model will perform in practice. One round of cross-validation involves partitioning a sample of
data into complementary subsets, performing the analysis on one subset (called the training
set), and validating the analysis on the other subset (called the validation set or testing set). To
reduce variability, multiple rounds of cross-validation are performed using different partitions,
and the validation results are averaged over the rounds.
K-fold cross-validation, the original sample is randomly partitioned into k equal size
subsamples. Of the k subsamples, a single subsample is retained as the validation data for
testing the model, and the remaining k-1 subsamples are used as training data. The crossvalidation
process is then repeated k times (the folds), with each of the k subsamples used
exactly once as the validation data. The k results from the folds then can be averaged (or
otherwise combined) to produce a single estimation. The advantage of this method over
repeated random sub-sampling is that all observations are used for both training and validation,
and each observation is used for validation exactly once. 10-fold cross-validation is commonly
used, but in general k remains an unfixed parameter.
5.1. K-Fold Cross-Validation
In stratified k-fold cross-validation 14, 15, the folds are selected so that the mean
response value is approximately equal in all the folds. In the case of a dichotomous
classification, this means that each fold contains roughly the same proportions of the two types
of class labels.
0
2
4
6
8
10
Fire Detected Not Detected
Fire
No Fire
0%
50%
100%
No Fire Fire Expected
Accuracy
Accuracy
? ISSN: 2502-4752
IJEECS Vol. 3, No. 3, September 2016 : 618 – 625
624
2 fold-cross validation, this is the simplest variation of k-fold cross-validation. For each
fold, we randomly assign data points to two sets d0 and d1, so that both sets are equal size (this
is usually implemented as shuffling the data array and then splitting in two). We then train on d0
and test on d1, followed by training on d1 and testing on d0. This has the advantage that our
training and test sets are both large, and each data point is used for both training and validation
on each fold.
Now 2 folding method applied on ten classes of image, that means 50 percent image of
a class are in training set and 50 percent images of that class are on test, that started 2 class, 3
class. The results are given on Table 3. Per class success and failed rate chart in Figure 5.
The next section that describes the accuracy rate of two classes and the expected class
in below chart Figure 6.

6. Conclusion
In this paper, image processing based fire detection system using color model was
proposed. We have collected a number of sequential frames from original video, which consists
of fire and non fire images. The proposed method consists five main stages: – collected Image
from video frame, convert the image to RGB color, selection of yellow color range of the image,
convert image to binary range, and detect the intensity of fire by contour binary image. The
proposed method is applied on video sequences and detected fire is classified into three groups
such as small fire, medium fire and large fire based on the threshold values.
7. Future Directions
To accomplish more valuable and more accurate video fire detection, this paper points
out future directions. We will improve the result for the step of intensity of fire by contour binary
image. Feature extraction method will be included before step five and we will use machine
learning algorithms like SVM, and KNN as a classifier to detect the fire more accurately by
replacing step five.
Acknowledgements
In this work I am grateful to Dr. Mohammad Khairul Islam, Professor Department of
Computer Science & Engineering, University of Chittagong, Bangladesh for his idea. I have had
inspired by him as well as the necessity of current world.