You are on page 1of 4

2014 Fifth International Conference on Intelligent Systems Design and Engineering Applications

Intelligent Fire Warning System Based on Fuzzy Extension Constraint Algorithm

Zhao Peng
(Department of Computer,Taiyuan Normal Uiversity ,Taiyuan, Shanxi Province,030012, China)
359896177@qq.com

Abstract—The traditional fire alarm systems sense the subject today. More and more people are paying attention to
temperature change produced fire and take the fire warning. research and improve the fire alarming and fire feature
When the space height is too high, and the gas velocity is too recognition technology[1].
big, all the environment factors influence the sensor working, Traditionally, the fire alarming system is established
and the induction effect is not good, the early warning cannot based on the temperature sensing and gas density change.
be reached. The intelligent fire warning system is proposed When the fire is not big, or the space height is too high, the
and designed based on fuzzy extension constraint algorithm, prevention and warning performance is not good. And
the detailed system design scheme is obtained. The short fire
recognition accuracy is low, and the prevention effect is not
feature recognition algorithm is presented, and in the
satisfying[2]. With the development of computer and signal
characteristics recognition process, the concept of feature
space transformation elastic restraint is proposed, the elastic
processing technology, the flam feature processing
fuzzy matching recognition region is established for the fire technology is developed, how to use the flame feature to
pixel feature, and the fuzzy recognition problem caused by recognize the fire is becoming an effective concept in fire
small flame features is solved. The intelligent fire detection alarming system. In this paper, the concept of feature space
and warning system is established, and the experiment is taken transformation elastic restraint is proposed, the elastic fuzzy
based on the system, the fire monitoring and detection result matching recognition region is established for the fire pixel
show that the new system can avoid the recognition defects feature, and the fuzzy recognition problem caused by small
when the flame of spark is small, the fuzzy color feature of flame features is solved. The intelligent fire detection and
flame also can be recognized perfectly. It provides a new way warning system is established, and the experiment is taken
for the design of intelligent fire detection system. based on the system, the fire monitoring and detection result
show that the new system can avoid the recognition defects
Keywords- Fire alarm; Fuzzy extension; Image feature when the flame of spark is small, the fuzzy color feature of
flame also can be recognized perfectly. It provides a new
way for the design of intelligent fire detection system.
I. INTRODUCTION
Fire disaster is a kind of burning phenomena out of II. FIRE INTELLIGENT DETECTION BASED ON FUZZY
control in time and space, and it is one of the major natural EXTENSION CONSTRAINT GAUSS FEATURE
disasters to human. In all sorts of disaster, fire is the most
frequent and most major disasters threatening public safety A. Establishment of fuzzy extension constrain feature
and social development, therefore, how to prevent fire is an model
important subject in research. In fact, fire not only destroy For the fire feature detection, the first step is to establish
the material property, cause social in disorder, and fire also the fuzzy extension constraint Gauss model[3], on the basis of
directly or indirectly endanger human people’s life, it takes the fuzzy extension constraint Gauss model, the flame region
great harm to people and society. Today, the fire disaster a and background region of fire image can be distinguished
common disaster problem all over the world must face. accurately. The characteristic ambiguity resulted from short
According to the statistics in China, it total number of fire scintillation time is overcome. In the process of fire
more than 200000 every five years, among them, the detection, the state of fire region changes in random. The fire
extraordinarily big fire disaster number more than 88, in feature extraction is difficult, and the gray feature extraction
every extraordinarily big fire, more than 50 households, or of flame changes in random[4]. The gray feature extraction of
direct property more than 1000000 Yuan, or killed more flame image should be improved in this paper, and the
than 10 people. Every day, 118 fires occurred in average, feature extraction algorithm is shown as follows:
and the death people is 18, 2378000 Yuan property are Assumed the fire is the electrical fire, and the state
burned. In 2013 for all over the world, the extraordinarily change number of electrical fire region is L, the state change
big fire is more than1091, more than 2172 people were can be described as the degree, the state change degree of
burned to death, and the direct economic loss is more than fire region can be expressed as:
2200000000 Yuan. From 2000 to 2013, China’s annual fire 1
  j ,u 1Yu vj ,u 
T
L xj,u Yu v j ,u
number is about 200000, and the casualties are more than QYu    p 1
e 2 (1)
2000 every year. The fire security situation is grim, and the
fire disaster takes great damage to people. In order to reduce
j 1
2   j,u
2 2

the fire occurrence, the prevention and warning is the key,


therefore, the research of the fire alarming and warning Where, x j ,u is the weight coefficient of j state change
technology and system is becoming an important and hot component in the u model, the weight coefficient should

978-1-4799-4261-9/14 $31.00 © 2014 IEEE 578


DOI 10.1109/ISDEA.2014.135
Yu  v j ,u  2.5 2j ,u , the feature matching is successful,
L
meet the conditions:  x j ,u  1
j 1
. And
otherwise, the matching is failure. Use the following formula
c T to update the number j sample data, the formula is:
Yu   y , y , y 
is the flame color component of the
s h

v j ,u  1  j ,u  v j ,u 1  j ,uYu
u u u
(5)
fire image in the moment. p is a constant, usually the value
j,u  1 j,u  2j,u1  j,u Yu  vj,u  Yu  vj,u  (6)
T T
is 3, and v j ,u   v sj ,u , v hj ,u , v cj ,u 
is the mean value of
j ,u
color component for the state model,  j ,u
 2j ,u J p is the Where, is the data update efficiency, it meets the

mean variance in the model[5]. condition: j ,u  ,  is the training time parameter
Here, the concept of fuzzy extension interval is x j ,u
introduced, using the following formula, the gray feature of sample, and the value is 0.38. Assumed there is no
component fuzzy extension constraint interval can be successful matching sample in the sample data set, the gray
calculated:
mean is set as Yu . The model of data is adjusted to the
n a  minimum weigh and maximum standard deviation. Use the
G(n , e )  exp   (2)
2 gy gz 1 2  21 2   following formula to update the state weights coefficients:
 
x j ,u  1    x j ,u 1   N j ,u (7)
Where, (  n ,  e ) is the gray characteristic interval,
Among them, assuming the sample j match
the interval is restricted by the constraint condition as
follows: Yu successfully, then value of N j ,u is 1, else the value is 0.
2
 n cos  e  g y   n cos  e  g y  2 Use the following formula for the flame region gray feature
a   2   ,  j ,k . extraction in the fire image:
 gy   gy  g y gz
    c 
In the above formula, the data can be obtained according C  arg max c   xL  u  , 0.3  u  0.8 (8)
to the mean and standard deviation gray distribution.  L 1 
Flame image gray feature distribution is shown in Figure
B. Extraction of flame gray feature 1.
In the establishment of a fuzzy feature space, the fuzzy
extension constraint Gauss feature model is used[6], and the
flame gray features of fire image are extracted. The
processing course includes Gauss model data preprocessing,
data update of Gauss model and gray feature extraction. The
specific steps are presented as follows:
The data in the model is preprocessed with filtering and
anti-interference technology, and we need to establish a set
of data with collected fire region image[7]. The sample data Figure 1. Flame image gray feature distribution
in the data set is ordered according to the time, the gray
degree mean is  0 , the gray standard deviation is 0 , the Through the above way, flame region gray feature of fire
image is obtained, and it provides the data basis for the fire
model data preprocessing is taken with the formula as
detection.
follows:
1 P 1 C. Intelligent fire recognition
0   Ju
P u 0
(3) According to the method described above, we can extract
gray feature of fire flame region in the image, and
1 P 1 registration recognition processing is taken for the image,
o    J u  0 
2
(4) and the fire detection is realized. The specific steps are
P u 0 expressed as follows:
Through the above algorithm, the model of data In the fire image, the gray means in different angles are
preprocessing is obtained, the following way is used to calculated, and the change state of gray means in different
explain fuzzy extension constraint Gauss model updating angles can reflect the scale and degree of gray change. Set
data processing. So, we should give the first fire image gray the size of fire image scanning window is 3  3 , and the
mean to the Gauss model, also the maximum variance and
gray mean variation coefficients of horizontal direction,
minimum weight coefficients are calculated. In dealing with
vertical direction and the diagonal direction are calculated by
the fire image, the color component Yu and sample data are the formulas as follows:
matched with feature correlation calculation. If

579
4 3 c=normxcorr2(sub_image1(:,:,1),sub_image2(:,:,1)); %
Wi  (Q( y  j, z  k )  Q( y  j 1, z  k ))2 (9) normalized cross correlation processing for the flame flicker
k 0 j 0 ribbons
3 4 figure;
Ww  (Q( y  j, z  k )  Q( y  j, z  k 1))2 (10) surf(c); %Display the threshold and normalized correlation
k 0 j 0 value
3 4
shading flat
We  (Q( y  j, z  k )  Q( y  j 1, z  k 1))2 (11) [max_c,imax]=max(abs(c(:)));
k 0 j 0
[ypeak,xpeak]=ind2sub(size(c),imax(1)); % Transform
3 4
W  (Q( y  j, z  k)  Q( y  j 1, z  k 1))2 (12) the one dimensional coordinates into two-dimensional
k 0 j 0
coordinate
corr_offset=[(xpeak-size(sub_image1,2)),(ypeak-
W ( y, z)  min(Wi ( y, z),Ww ( y, z),We ( y, z),W ( y, z)) (13) size(sub_image1,1))];
Use the following formula to calculate the image rect_offset=[(rect_image2(1)-
matching gradient coefficient: rect_image1(1)),(rect_image2(2)-rect_image1(2))];
E(v, w)   x( y, z) J ( y  v, z  w)  J ( y, z) (14) offset=corr_offset+rect_offset;
2
xoffset=offset(1); % offset value of x axis direction
y, z
yoffset=offset(2); % offset value of y axis direction
Where, x( y, z ) image matching window, xbegin=round(xoffset+1);
 J ( y  v, z  w)  J ( y, z ) xend=round(xoffset+size(image1,2));
2
is the mage gray difference ybegin=round(yoffset+1);
'
value, I ( y , z ) is the Gauss model window. Use the yend=round(yoffset+size(image1,1));
extracted_ image1=image2(ybegin:yend,xbegin:xend,:);
following formula to calculate the flame gray parameters in %Extract image1subfeature image
fire region.
if isequal(image1,extracted_image1)
v 
F (v, w)   v, w N  
disp(' image1.bmp was extracted from image2.bmp'); %
(15) To determine whether there is an association between two
w
images, and then determine whether there is fire hazard
Where end
 J2 JyJz 
N   x( y, z )  y  (16)
E. Simulation results and analysis
y,z  J y J z J z2
 For verifying the effectiveness of the fire alarming
system proposed in this paper, the simulation is taken in
The matching processing is taken for the Gray feature of experiment, and the simulation is taken based on 2 video
fire image, assumed the matching is successful, the detected sequences. The simulation process and results are shown as
fire is happening, else, the area does not exist hidden danger follows.
of fire. Finally, the fire detection and warning algorithm Video series 1: The video is collected from Internet site
design is completed. http://www.openvisor.org. The video has 1835 frames, and
D. Key source codes of program design the frame size is 320 * 240, the frame rate is 25fps. In the
beginning of the Video scene, there is no moving object, and
On the basis of MATLAB programming language, the
in the 262nd frame, there is a human enter into the video and
system software program is designed, and the source codes emit smoke in the 454th frame. The system gives the
of the fire recognition process are presented as follows: warning signal in the 535th frame, in the 569th frame, the fire
image1=imread(' image1.bmp'); %Read state is confirmed. It means that the system make the correct
image1 judgment in 5 seconds. Throughout the video, the wearing
image2=imread(' image2.bmp'); % Read clothe of the people is similar to the color of gray smoke,
image2 and walking around in the video as the interference. The
figure;subplot(121); environment is complicated, but the system hasn’t done the
imshow(image1); %Compare image1 misjudgment of the situation. The fire alarming is accurate
subplot(122); and timely. The simulation results for the video 1 are shown
imshow(image2); % Compare image2 in Figure 2.
rect_image1=[111 33 65 58]; %Determine image1 region
rect_image2=[163 47 143 151];% Determine image2 region
sub_image1=imcrop(image1,rect_image1);
sub_image2=imcrop(image2,rect_image2);
figure;subplot(121);
imshow(sub_image1);
subplot(122);
imshow(sub_image2);
(a) First frame (b) 290th frame

580
(c) 702nd frame (d) 987th frame

Figure 2. Fire detection and alarming for Video 1

Video series 2: The video consists of 5975 frames, the Figure 4. Missing judgment video
frame size is 320 × 240, and the frame rate is 25fps. Firstly,
there is no human in the Video scene, in the 75th frame, the
people appeared, and hold a red bucket with gasoline, and III. CONCLUSIONS
the red color of the bucket is similar to the flam color. The In this paper, the intelligent fire warning system is
human keeps staying in the image till 1579th frames. In this proposed and designed based on fuzzy extension constraint
course, the system does not misjudge the situation. In the algorithm, the detailed system design scheme is obtained.
1855th frame, the person returns to the site, and in the 2311th The short fire feature recognition algorithm is presented, and
frame, the people burns out the firewood and the gas is in the characteristics recognition process, the concept of
breaking out, in the 2402nd, fire occurred in the 2402nd feature space transformation elastic restraint is proposed, the
frame, and burn out rapidly. The system gives the warning elastic fuzzy matching recognition region is established for
alarming in 2464th frames, in 2488th frame, the fire station is the fire pixel feature, and the fuzzy recognition problem
confirmed. It means that the system make the correct caused by small flame features is solved. The intelligent fire
judgment in 3.44 seconds. The whole process is shown in detection and warning system is established, and the
figure 3. experiment is taken based on the system, the fire monitoring
and detection result show that the new system can avoid the
recognition defects when the flame of spark is small, the
fuzzy color feature of flame also can be recognized perfectly.
For most of the fire video, the system can make the correct
judgment in 8 seconds, it provides a new way for the design
of intelligent fire detection system and it has good
(a) First frame (b) 1579th frame application value in fire warning and alarming system design.

REFERENCES
[1] ZHANG Yi, SHENG Huiping, HU Guangbo. Study on Compressor
Fault Diagnosis Based on Space Reconstruction and K-L
Transform[J].Compressor Technology. 2011, 4: 19-21. (in Chinese)
[2] CAO Jian, LI Hai- sheng, CAI Qiang. Research on Feature
Extraction of Image Target[J]. Computer Simulation. 2013; 30(1):
(c) 2311th frame (d) 2464th frames 409-413. (in Chinese)
[3] Zheng Baodao, Wang Huaijie. Design of Remote Image Alarming
Figure 3. Fire detection and alarming for Video 2 System Based on ARM/GPRS. Computer Measurement & Control.
2013; 21(1): 149-151, 159. (in Chinese)
For most of the video, the system can make the correct [4] LUO Hai-chi, LI Yue-Yang, SUN Jun. Filtering Method for Images
Based on Adaptive Neuro-fuzzy Interference System[J]. Computer
judgment in 8 seconds (Frame rate is 25fps), and the real- Science. 2013; 40(7): 302-306. (in Chinese)
time property of the system is very high, the experimental [5] Zhao Juan, Zhao Qiang, Wu Fenxia. Self-organizing Map Neural
platform in this paper can reach 15 ~ 30fps, the computing Network Based Image Retrieval Algorithm[J]. Bulletin of Science
cost is low. Among them, there are two video appeared and Technology. 2013; 29(2): 55-57. (in Chinese)
missing judgment, the reasons are the occupation resource is [6] MO Shao-qiang, LIU Jiang-lin. Design of Logistics Information
Collection System Based on Diagram Identification Technology[J].
too far, and the target is too small. The dynamic feature Logistics Technology. 2012, 31(9): 400-401, 416. (in Chinese)
extraction errors happened. And the other reason is that the [7] SUI Xin. Study and Implementation on Host Characteristic
smoke from coming out to appear is too short. But the fire Information Passively Identifying[J]. Science Technology and
warning is given out, and the fire monitoring and detection Engineering. 2013; 13(3): 652-658. (in Chinese)
result show that the new system can avoid the recognition
defects when the flame of spark is small, the fuzzy color
feature of flame also can be recognized perfectly. It provides
a new way for the design of intelligent fire detection system.
The missing judgment video is shown in figure 4.

581

You might also like