An automatic fire searching and suppression system for large spaces

ARTICLE IN PRESS
Fire Safety Journal 39 (2004) 297–307
An automatic fire searching and suppression
system for large spaces
Tao Chen, Hongyong Yuan*, Guofeng Su, Weicheng Fan
State Key Laboratory of Fire Science, University of Science and Technology of China,
Hefei 230027, PR China
Received 25 November 2002; received in revised form 11 August 2003; accepted 19 November 2003
Abstract
An automatic fire searching and suppression system with remote-controlled fire monitors
for large spaces is developed. The fire searching method is realized based on computer vision
theory via one CCD camera fixed at the end of a fire monitor chamber. While the fire monitor
is pivoting, continuous images are taken from the CCD camera and transmitted into a
computer. Then the images are processed with an image fire detection method to determine
whether a fire occurs in the vision field. Once a fire is detected, it will be automatically located
through an image processing algorithm. Displacement and pivot angle of the CCD camera in
searching process are the essential parameters to calculate the space coordinates of a fire.
When the coordinates of the fire are obtained, the fire monitor can be adjusted to the
appropriate direction and elevation to spray according to water pressure.
r 2003 Elsevier Ltd. All rights reserved.
Keywords: Automatic fire monitor; Fire searching; Fire suppression; Large spaces
1. Introduction
Water is most acceptable for fire suppression in large space when performance and
quantity are taken into account. In this case, water type fire suppression apparatus
have many advantages such as larger protection area and low suppression cost.
Furthermore, it does not produce toxic products. Sprinklers, as the most common
suppression apparatus, have been widely installed for fire protection. However in
*Corresponding author. Fax: +86-551-3606430.
E-mail address: yuanhy@ustc.edu.cn (H. Yuan).
0379-7112/$ - see front matter r 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/j.firesaf.2003.11.007

ARTICLE IN PRESS
298
T. Chen et al. / Fire Safety Journal 39 (2004) 297–307
Nomenclature
D chamber length of the fire monitor
d measured distance between a fire and the pivot point of the fire monitor
d c calculated distance between a fire and the pivot point of the fire monitor
f focal length of the CCD camera
e distance between a fire and the shooting spot of the fire monitor
(i 1 , j 1 ) pixel coordinates of a fire in the image at time t 1
(i 2 , j 2 ) pixel coordinates of a fire in the image at time t 2
H maximum pixel number per row of the CCD
h actual width of the CCD
T time of extinguish after the water is issued
T s time of searching process
V maximum pixel number per column of the CCD
v actual height of the CCD
X–Y–Z fixed object space coordinate system
X 1 –Y 1 –Z 1 camera coordinate system at time t 1
X 2 –Y 2 –Z 2 camera coordinate system at time t 2
(X,Y,Z) coordinates in X-Y-Z coordinate system
(X 1 ,Y 1 ,Z 1 ) coordinates in X 1 -Y 1 -Z 1 coordinate system
(X 2 ,Y 2 ,Z 2 ) coordinates in X 2 -Y 2 -Z 2 coordinate system
x–y image plane of the camera
(x,y) coordinates in x–y image plane
(x 1 ,y 1 ) coordinates in image plane at time t 1
(x 2 ,y 2 ) coordinates in image plane at time t 2
a angle to Z-axis in Y–Z plane of the fire spot
b visual angle of the CCD camera
o 1 angle to Z-axis in Y–Z plane of Z 1 -axis
angle to Z-axis in Y–Z plane of Z 2 -axis
o 2
recent years, with the increasing large space buildings, the suitability of sprinkler use
in such large space situations has been challenged.
Commonly, thermally activated sprinklers are not suitable for high ceiling space
use. The height of the ceiling in large space will greatly affect the sprinklers so that
they cannot provide effective protection. The main problem is the activation delays
of the sprinklers in large spaces, particularly high ceiling spaces. For example, even a
5 MW fire on the atrium floor would not generate a smoke temperature high enough
to activate the sprinklers on an atrium ceiling higher than 15 m [1, 2]. Due to the
distance between the potential fire source and the sprinkler heads which are mounted
near the ceiling, sprinkler activation may be too late to minimize the fire hazards or
damage. Even the early suppression fast response (ESFR) sprinklers have the
recommended height limit [3–5]. Moreover, the height may also affect how
much water can reach the flame before vaporized or blown away by the fire
plume [6]. In those large spaces where the combustible materials are distributed

ARTICLE IN PRESS
T. Chen et al. / Fire Safety Journal 39 (2004) 297–307 299
dispersedly, it is also not very efficient to install many sprinklers to cover all these
materials.
Now, robotic fire monitor that can be automatically controlled by computer and
can work as a linkage device in a fire detection and suppression system is becoming a
more suitable solution. Liu et al. [7] and Yuan et al. [8] have done some early
researches. Compared to a sprinkler, a fire monitor has many advantages. First, the
most remarkable advantage is that the activation time of these fire monitors is much
shorter than sprinklers. They are activated by the alarm signal given by the fire
detectors which are much more sensitive than the sprinklers’ explosion glass bulbs.
These detectors can be conventional fire detectors or gas sensors. Second, water
stream from a fire monitor has a higher speed, a greater flux and a larger impulse so
it is more effective for fire suppression. Its spot type suppression may sometimes
avoid the consequent losses caused by sprinklers and will not cause a downward
moving of smoke [9] to endanger the occupants in the building. Last, a typical fire
monitor has a shooting range of 50 m long, which enables it to protect much larger
area than a sprinkler.
In this paper, a new fire searching and suppression system using such automatic
fire monitors is described. These fire monitors are assembled into a larger fire
detection–suppression system. After fire confirmation and searching process, the
direction and elevation of the fire monitor can be easily calculated and suppression
can consequently be executed.
2. Principle
2.1. Structure of the automatic fire monitor system
The automatic fire monitor system is a fire detection–searching–suppression
system [10] which consists of three parts: the fire detection module (including
detectors and the detector controller), the fire monitor module (including the fire
monitor, the fire monitor manual controller and the CCD camera) and the central
control module (including the computer, data-sampling interfaces and communication
interfaces). In this system (shown in Fig. 1), the fire monitor serves as a fire
searching and suppression device in conjunction with the fire detectors. Once the
detector is in alarm state, the fire detection module will give a fire alarm signal and
the central control module will send a linkage signal to activate the corresponding
fire monitor.
The fire monitor module and the central control module accomplish the fire
searching and suppression after the fire monitor is activated. An automatic fire
monitor itself is much like a little barbette which consists of a chamber, a CCD
camera, a motor, an electromagnet valve, a computer connection interface (RS485
and video) and a shell. At the end of the fire monitor chamber, a CCD camera is
fixed which is used for capturing the spot images. The optical axis of the CCD
camera is adjusted parallel to the axis of the fire monitor chamber. In large space
condition, the two axes can be practically considered coincident. The electromagnet

ARTICLE IN PRESS
300
T. Chen et al. / Fire Safety Journal 39 (2004) 297–307
Fig. 1. Sketch of automatic fire monitor system.
Table 1
Technical specification of the water cannon
Maximum operating pressure
16 bar (16 10 5 Pa)
Operating pressure
8 bar (8 10 5 Pa) (lower pressure is allowable)
Flow rate
1200 lpm
Pivot range
80 ?+60 max vertical, 7100 max horizontal
Pivoting speed 9 /s
Remote control
RS-485 bus
Computer interface
RS-232 (with a protocol conversion box)
valve and the motor inside the fire monitor shell are controlled by the computer via
RS-485 serial communication interface. The former releases or holds the water jet
and the latter drives the fire monitor chamber to pivot in left–right and up–down
directions. Furthermore, the pivot angle of the chamber is fed back to the computer
as an important parameter to the fire searching algorithm.
The fire monitor has dual control modes. A manual fire monitor controller is used
for manual operations at emergency situation. In case a fire is discovered by a
human, the cannon can be operated manually by a trained personnel. In the
automatic mode, multiple fire monitors can be controlled and switches between them
are available. The main technical specifications of a typical fire monitor are shown in
Table 1.
2.2. Fire searching method based on computer vision theory
Computer vision tries to simulate man’s vision and reconstruct 3D scene with 2D
images. Usually, at least two cameras are needed to construct a 3D vision system
[11]. Baseline distance between the two cameras is essential to reconstruct 3D
coordinates. The relationshipbetween image coordinates and object coordinates

ARTICLE IN PRESS
T. Chen et al. / Fire Safety Journal 39 (2004) 297–307 301
based on single fixed camera imaging system can be established as
x i ¼ f X o
Z o
;
y i ¼ f Y o
Z o
;
ð1Þ
where (x i , y i ) represent the image coordinates, (X o , Y o , Z o ) represent the object
coordinates and f represents the camera’s focal length.
Here we use only one CCD camera to construct a vision system according to the
position change of the camera between time t 1 and t 2 when it is pivoting in Y–Z
plane. Fig. 2 shows the coordinates system in our vision system.
* The pivoting point of the camera–fire monitor system is the origin of X–Y–Z
coordinate system.
* The camera’s optical center is the origin of the X 1 –Y 1 –Z 1 and X 2 –Y 2 –Z 2
coordinate systems.
* X–Y–Z, X 1 –Y 1 –Z 1 and X 2 –Y 2 –Z 2 are right-handed coordinate systems.
* In X–Y–Z coordinate system, o measures zero along Z-axis and positive by righthand
rule with X-axis as the thumb direction. Thus we have o 1 o0 and o 2 >0.
* P(d,a) in the Y–Z plane is supposed to be the fire position in object space.
From Eq. (1) we have
Z 1 ¼ f Y 1
y 1
;
Z 2 ¼ f Y 2
y 2
:
ð2Þ
Between object space coordinates and image space coordinates we have
Y ¼ Y 1 cos o 1 Z 1 sin o 1 þ D sin o 1 ;
Z ¼ Y 1 sin o 1 þ Z 1 cos o 1 þ D cos o 1 ;
ð3Þ
Fig. 2. Coordinates built from computer vision theory.

ARTICLE IN PRESS
302
T. Chen et al. / Fire Safety Journal 39 (2004) 297–307
Y ¼ Y 2 cos o 2 Z 2 sin o 2 þ D sin o 2 ;
Z ¼ Y 2 sin o 2 þ Z 2 cos o 2 þ D cos o 2 :
ð4Þ
Take Eq. (2) into (3) and eliminate Y 1 , the relationshipbetween Y and Z has an
expression
Y ¼ ðy 1 cos o 1 f sin o 1 ÞðZ D cos o 1 Þ
þ D sin o 1 : ð5Þ
y 1 sin o 1 þ f cos o 1
Take Eq. (2) into (4) and eliminate Y 2 , the relation between Y and Z has another
expression
Z ¼ ðy 2 sin o 2 þ f cos o 2 ÞðY D sin o 2 Þ
þ D cos o 2 : ð6Þ
y 2 cos o 2 f sin o 2
Solution of Y and Z can be obtained from Eqs. (5) and (6), as shown in Eq. (7).
Y ¼ Dðy 2 f tan o 2 Þ½ðsin o 2 sin o 1 þ A cos o 1 B cos o 2 Þ=ðA þ BÞ D cos o 2 Š
y 2 tan o 2 þ f
þ D sin o 2 ;
Z ¼
D
A B ðsin o 2 sin o 1 þ A cos o 1 B cos o 2 Þ; ð7Þ
where
A ¼ðy 1 f tan o 1 Þ=ðy 1 tan o 1 þ f Þ;
B ¼ðy 2 f tan o 2 Þ=ðy 2 tan o 2 þ f Þ: ð8Þ
In the above equations, y 1 and y 2 are still unknown. Actually, they can be
calculated from the pixel coordinates (i, j) of the images, because most of the CCD
camera are of the same standard specifications. Suppose the CCD camera’s scanning
area is known as H V pixels and h v m 2 , the relationshipbetween [y 1 , y 2 ] and
[j 1 , j 2 ] can be found.
y 1 ¼ðj 1 V=2Þv=V;
y 2 ¼ðj 2 V=2Þv=V: ð9Þ
Thus the fire position P(d,a) in the space is confirmed and expressed as
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
d ¼ Z 2 þ Y 2 ; ð10Þ
a ¼ tan 1 ðY=ZÞ:
ð11Þ
2.3. Working process
Initially, the fire monitor is set at the horizontal-left limit position. Once a fire is
detected by a fire detector in the system, the corresponding fire monitor will be
activated. It will start searching for fire and be driven to the down-right limit
position step-by-step (shown in Fig. 3). The fire monitor rests for a short time after
each step. During this period of time, continuous images are captured from the CCD

ARTICLE IN PRESS
T. Chen et al. / Fire Safety Journal 39 (2004) 297–307 303
Fig. 3. Fire monitor’s scanning process. 1: horizontal-left limit; 6: down-right limit.
I
II
t 1
y
fire monitor pivot
fire monitor pivot
x
III
t 2
fire monitor pivot
IV
Fig. 4. Aiming process when a fire is detected.
camera and sent into computer to be processed by image fire detection algorithm [12].
In the searching process, the robotic fire monitor scans in horizontal direction for
three times however, each time in a different vertical angle. If no fire is detected in the
entire searching process when the fire monitor reaches its down-right limit, it will be
reset to its initial position. Otherwise, the locating algorithm is employed.
In the locating algorithm, the distance between the fire monitor and fire is
calculated. The stepangle of the fire monitor is chosen as b (less stepangle can be
chosen) which is the visual angle of the CCD camera, thus no protection area is
skipped. By pivoting the fire monitor, we perform the calculation by the
displacement at different time. Images I, II, III and IV in Fig. 4 show the searching
and aiming movements.

ARTICLE IN PRESS
304
T. Chen et al. / Fire Safety Journal 39 (2004) 297–307
* First, when a fire region is detected on the image, mark the region and its gravity
center. Move the fire monitor and mark the fire region repeatedly until the fire
region reach the bottom edge and x-center of the image.
* Second, the vertical pivot angle of the fire monitor at this time (t 1 ) is recorded as
o 1 . Move the fire monitor vertically until the fire region reach the topedge of the
image.
* Third, the vertical pivot angle at this time (t 2 ) is recorded as o 2 . Move the fire
monitor vertically until the fire region is at the center of the image. Thus
combined with steptwo, the coordinate systems as shown in Fig. 2 are formed.
* Last, now the fire monitor nozzle is pointed right to the fire, record the vertical
pivot angle as o 3 .
Now the distance d between the fire and the fire monitor can be calculated from
computer vision theory with the aforesaid coordinate systems. Longer baseline yields
more accuracy in fire location, so in image II and image III, the fire region is moved
from one edge to another so that |o 2 o 1 | is as large as possible to achieve long
baseline.
2.4. Suppression characteristic
Above describes the fire searching process. To shoot the fire with water stream,
change the elevation of the chamber according to d, a, o 3 and water pressure. After
the water pump is started up, the fire monitor can spray once the electromagnet valve
is opened. While spraying, the chamber will swing in about 75 in all directions to
fully cover the fire region. The diameter of water stream can be adjusted according to
distance and environment in order to achieve best suppression effect.
The maximum detection distance of the CCD is 100 m and the maximum spray
distance of the fire monitor is 50 m. The protection area is considerable for the
application in large space. By proper installation of multiple fire monitors, the
protection area can be almost fully covered. Fig. 5 shows the protection area. The
vertical limit of the fire monitor is 80 , so there will be a dead area right below each
fire monitor. Usually the dead area can be ignored because it is relatively small. For
example, a fire monitor mounted at 8 m only has a dead area of about 3.1 m 2 .
Fig. 5. Protection area of the fire monitors.

ARTICLE IN PRESS
T. Chen et al. / Fire Safety Journal 39 (2004) 297–307 305
3. Experiment
Experiments were conducted in a large test hall which is 30 m high, 30 m long and
18 m wide. The fire monitor was installed on one wall at the height of 6 m. Fuels in
different tests were placed in a line on the ground of the hall every 2 m from the fire
monitor (recorded as L). In each test, the fire monitor started its search from the
horizontal-left limit. Water pressure is 4 bar. Different fuels, such as diesel oil, wood,
and paper box are tested separately. Fig. 6 shows the experimental setup.
Usually, the spraying lasts for 10 s and during this time the chamber will swing in
75 in all directions for 3 times.
Table 2 shows the data for diesel oil fire tests. The fire is a 0.2 m 0.2 m pool fire
ignited by flock.
In all tests, the values of e are obtained by eyeballing and are of low precision.
Searching time T s varied in different tests, but usually it was less than 30 s if a fire is
found. In the test data of the same fire type, all the former three T s values were
obviously greater than the later three T s values. This was caused by the limitation of
the CCD camera’s vision field since the fire monitor started its search from the same
point in our tests. Fires farther than 16 m were detected in 1–2 scanning period
(reference to Fig. 3) and fires between 10 and 16 m were detected in 3–4 scanning
period (reference to Fig. 3) after the fire monitor chamber pivoted down. Generally,
Fig. 6. Experimental setup.
Table 2
Experimental data (diesel oil)
Number Fire type L (m) d=(h w 2 +L 2 ) 0.5 (m) d c (m) T s (s) e (m) T (s)
1 Diesel oil 10.0 11.66 11.86 25.3 0.3 5.1
2 Diesel oil 12.0 13.42 13.14 22.2 0.1 3.2
3 Diesel oil 14.0 15.23 15.07 23.8 0.2 4.0
4 Diesel oil 16.0 17.09 17.26 14.9 0.1 3.7
5 Diesel oil 18.0 18.97 18.93 12.2 0.0 2.9
6 Diesel oil 20.0 20.88 20.30 12.3 0.1 4.2

ARTICLE IN PRESS
306
T. Chen et al. / Fire Safety Journal 39 (2004) 297–307
searching time varied with fire position, fire strength and computation depth of our
detection algorithm, which will not be further discussed here.
Calculated distance and the actual distance are fairly accordant. The deviation is
mainly due to the low precision of the pivot angle fed back to the computer. The
angle measuring device in the fire monitor is quite simple and the motor gears also
have mechanical errors. Extinguishing time T is usually smaller than 6 s because the
fire strength in our tests is very small while the water flux is relatively large.
Generally, larger e caused longer extinguishing time. Usually, if the water stream
dose not hit the fire spot initially, it will cover the fire spot while the fire monitor is
swinging.
4. Conclusion
Fire searching and suppression were combined together and realized automatically
in the automatic fire monitor system. Displacement and pivot angle of the CCD
camera in fire searching process are the essential parameters to calculate the space
coordinates of fire. Once the direction and the distance of the fire are obtained, the
horizontal pivot angle and the vertical elevation of the fire monitor can be calculated
according to water pressure. The fire searching precision is enough for practical use
and the suppression effect is also satisfactory.
Remote and automatic control of the fire monitor improves its efficiency and
adaptability. Moreover, it provides more safety in fire fighting and is very
economical when incorporated with fire protection system for large spaces use.
Instead of installing multiple layers of sprinklers, the application of the fire monitor
will not influence the inside sight of the buildings. It can be very useful to apply this
technology in large spaces such as large museums, palaestras, showplaces and
storehouses.
References
[1] Chow WK. Performance of sprinkler in atria. J Fire Sci 1996;14(6):466–88.
[2] Gott JE, Natariani KA. Analysis of high bay hangar facilities for detector sensitivity and placement.
Society of fire protection engineers. Honors Lecture Series, May 20, 1996, Proceedings. Engineering
seminars: Fire Protection Design for High Challenge of Special Hazard Applications, May 21–22,
1996, Boston, MA, 1996.
[3] Hankins J. ESFR sprinklers–the facts you should know. Fire Prevention and Fire Safety Eng J 1995;
2(6).
[4] Kung HC, et al. Early suppression fast response (ESFR) sprinkler protection for 12 m high
warehouses. Proceedings of the Fifth International Symposium on Fire Safety Science. Melbourne,
1997.
[5] Yao C. Overview of sprinkler technology research. Proceedings of the Fifth International Symposium
on Fire Safety Science. Melbourne, 1997.
[6] Chow WK. Full-scale burning facilities for atrium fire research. Fire Safety Sci 1993;2(2):61–4
[in Chinese].

ARTICLE IN PRESS
T. Chen et al. / Fire Safety Journal 39 (2004) 297–307 307
[7] Liu Shenyou, et al. Automatic orientating fire extinguishing monitor. China Safety Sci J 2001;
11(2):37–41 [in Chinese].
[8] Yuan Hongyong, Chen Tao, Su Guofeng, Liu Binghai. Fire location and suppression with automatic
hydrant in large space. Second NRIFD Symposium—Science, Technology and Standards for Fire
Suppression Systems, Proceedings. Tokyo, Japan: 17th–19th July, 2002.
[9] Cooper LY. The interaction of an isolated sprinkler spray and a two-layer compartment fire
environment. Phenomena and model simulations. Fire Safety J 1995;26(1):1–33.
[10] Fan Weicheng, Yuan Hongyong, Liu Shenyou. The integrated system for fire detection and
extinction. Fire Safety Sci 1995;4(3):49–53 [in Chinese].
[11] Gao Wen, Chen Xilin. Computer vision—algorithm and principle. Tsinghua University Press:
Beijing; 1999.
[12] Yuan Hongyong, et al. Image type intelligent fire detection and location method. Safety Eng 1996;
(3):17–19 [in Chinese].