See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/305190416

A SURVEY ON FOREST FIRE DETECTION

Article · January 2015

DOI: 10.16962/EAPJCSIS/issn.2394-0441/20150930.V1I2.03

CITATIONS READS

2 5,924

2 authors, including:

L. Latha
Kumaraguru College of Technology
17 PUBLICATIONS 182 CITATIONS

SEE PROFILE

All content following this page was uploaded by L. Latha on 06 September 2018.

The user has requested enhancement of the downloaded file.

 ELK ASIA PACIFIC JOURNAL OF COMPUTER SCIENCE AND INFORMATION SYSTEMS

ISSN: 2454-3047 (Print) ISSN: 2394-0441 (Online) Volume 1 Issue 2 (2015)

www.elkjournals.com
……………………………………………………………………………………………………………………

A SURVEY ON FOREST FIRE DETECTION

P. Tamil Mathi Dr. L. Latha

PG Student ASP
Department of Computer Science and Engineering, Department of Computer Science and Engineering,
Kumaraguru College of Technology, Coimbatore, Kumaraguru College of Technology, Coimbatore,
India India

ABSTRACT


Fire detection at an early stage is important for the safety of the people. Lack of information due to manual
detection is the main cause of failure of fire detection. Fire can be detected by using smoke at an early stage as it is
the fire indicator. Generally automatic forest fire detection using image processing techniques represents one of the
significant aspects of forest fire avoidance earlier. Detection using image and video is effective than using sensors.
In image processing the inputs for the fire detection may be an image or a video but the input as a video is quite
complex process but provides good result. The techniques such as Wavelet decomposition, spatial and temporal
analysis, Gaussian Mixture Model, Multi-Feature fusion detect fire in an accurate manner.

Keywords: Wavelet and Color model, Static and Dynamic characteristic analysis, Wavelet Transform, Spatio-
temporal flame Modeling, Texture Analysis, Neural Network, Covariance Descriptors, Probabilistic approach

1. Introduction

Wildfires are a significant hazard to

ecological systems around the world and Different image processing techniques can
pose a serious threat to human safety. be used to detect fire and smoke. In image
People visually look for signs of fire or processing image or video is taken as input
smoke appearance to detect fire in older and the output may either an image or
days. To detect fire at early stage we use parameters or characteristics of an image.
smoke which is the good indicator of fire Various tasks like analysis classification,
which is visible before flames. extracting the features, recognizing
Characteristics of smoke need to be different patterns can be performed using
considered such as transparency, its image processing. The features and textures
response to environmental condition, its of smoke can extract using various image
shape. In open environment smoke processing techniques. By using these
detection pose a serious challenge in such techniques the dangerous situations caused
areas sensors may be used but this has due to fire can be avoided and safety of the
limitations such as time and wide area people can be preserved.
coverage. To overcome this video fire
detection systems are used.

32
 ELK ASIA PACIFIC JOURNAL OF COMPUTER SCIENCE AND INFORMATION SYSTEMS

ISSN: 2454-3047 (Print) ISSN: 2394-0441 (Online) Volume 1 Issue 2 (2015)

……………………………………………………………………………………………………………………

2. Literature Survey unwanted distortions are removed and

image is resized and transformation of
Surapong Surit, Watchara Chatwiriya [8] resized image is performed. High
proposed a method to detect fire by smoke frequencies of an image are eliminated
detection in video. This approach is based using SWT and the reconstruction of image
on digital image processing approach with is done by inverse SWT. Image indexation
static and dynamic characteristic analysis. is performed to group the intensity colors
The proposed method is composed of that are closed to each other. Histogram
following steps, the first is to detect the area analysis is used to determine the various
of change in the current input frame in levels of indexation. After analysis a
comparison with the background image, the comparison is made with non-smoke frame
second step is to locate regions of interest and non-smoke images are eliminated.
(ROIs) by connected component algorithm, These three are combined and fire is
the area of ROI is calculated by convex hull detected.
algorithm and segments the area of change
from image, the third step is to calculate Osman Gunay and Habiboglu [4] proposed
static and dynamic characteristics, using a system based on Covariance Descriptors,
this result we decide whether the object Color Models, and SVM Classifier. This
detected is the smoke or not. The result system uses video data. Spatio-temporal
shows that this method accurately detects Covariance Matrix (2011) [13] is used in
fire smoke. this system which divides the video data
into temporal blocks and computes
P. Piccinini, S. Calderara, and R. Cucchiara covariance features. The fire is detected
[2] proposed a method based on the wavelet using this feature. SVM Classifier is used to
model and a color model of the smoke. The filer fire and fire-like regions. This system
proposed method exploits two features: the supports only for clear data not for blur
variation of energy in wavelet model and a data.
color model of the smoke. Smoke is
detected based on the decrease of energy Dimitropoulos (2015) [1] proposed an
ratio in wavelet domain between algorithm where a computer vision
background and current. The deviation of approach for fire-flame detection is used to
the current pixel color is measured by the detect fire at an early stage. Initially,
color model. Bayesian classifier is used to background subtraction and color analysis
combine these two features to detect smoke. is used to define candidate fire regions in a
frame and this approach is a non-parametric
R.Gonzalez proposed a method to detect model. Following this, the fire behavior is
fire based on Wavelet Transform. modeled by employing various Spatio-
Stationary Wavelet Transform is used to temporal features such as color probability,
detect Region of Interest. This method flickering, spatial and spatiotemporal
involves three steps preprocessing, SWT, energy. After flame modeling the dynamic
histogram analysis. In preprocessing texture analysis is applied in each candidate

33
 ELK ASIA PACIFIC JOURNAL OF COMPUTER SCIENCE AND INFORMATION SYSTEMS

ISSN: 2454-3047 (Print) ISSN: 2394-0441 (Online) Volume 1 Issue 2 (2015)

……………………………………………………………………………………………………………………

region using Linear Dynamical Systems, phase. The first phase involves
Histogram and Mediods. LDS is used to segmentation of image using LBP. LBP is a
increase the robustness of the algorithm by texture operator whose value is computed
analyzing temporal evolution of pixel using image’s center and neighboring pixel
intensities. Pre-processing is done after this values. Further accuracy is improved using
to filter non-candidate regions. Spatio- Wavelet Transform and complicated data is
temporal analysis is done to increase the classified using this approach. 2D Discrete
reliability of the algorithm. The consistency Wavelet Transform is used for
of each candidate fire region is estimated to decomposition in this system. 2 images
determine the existence of fire in should be used as input and the sub bands
neighboring blocks from the current and of every image are compared with the
previous video frames. Finally, a two-class other, if sub bands are equal the images are
SVM classifier is used to classify the fire same else different.
and no fire regions.
(Refer Figure 2)
(Refer Figure 1)

Hamed Adab [6] proposed another system Celik (2007) [3] proposed a generic model
which is based on Indexing. GIS techniques for fire and smoke detection without the use
and remote sensing [10] provides further of sensors [15]. Fuzzy based approach is
assistance. The indexing may be structural used in this system. Color models such as
fire index, Fire risk index, Hybrid fire YCbCr, HSV are used for fire and smoke
index. Depending on the geographical detection. The fire is detected using YCbCr
condition of the area the indexing differs. color model samples because it
Validations of indices are based on hot spot distinguishes luminance and chrominance.
data. Structural fire indices show static Y, Cb, Cr color channels are separated from
information and it does not change over RGB input image. A pixel is more likely a
short time span and used to predict the risk fire pixel if intensity of Y channel is greater
in advance. Fire risk index changes as the than channel Cb and Cr.
vegetation or climate changes. Hybrid
index is a combination of Structure and Fire For example consider the following image
index. The disadvantage of this indexing is set,
that way of combining.
(Refer Figure 3)
Akshata & Bhosale [7] proposed another
method where Local Binary Pattern acts as In the above image set, the pixel is fire
a base for fire detection and Wavelet pixel as the intensity of Y channel is greater
Decomposition is used to detect fire. Pixel than Cb and Cr channel. HSV color model
level analysis is required in this method. is used for Smoke detection as is does not
This method uses YCbCr color model to show chrominance characteristics as fire.
detect fire. Detection is based on three As smoke is the early indicator of fire it

34
 ELK ASIA PACIFIC JOURNAL OF COMPUTER SCIENCE AND INFORMATION SYSTEMS

ISSN: 2454-3047 (Print) ISSN: 2394-0441 (Online) Volume 1 Issue 2 (2015)

……………………………………………………………………………………………………………………

should be detected at lower temperature, is used here. The number of neurons in the
here its color varies from white-bluish to output layer is equal to the number of
white, the saturation is low which satisfies desired parameters of the output vector,
the HSV color model property. As like which are “smoke,” “sky,” and
smoke, sky also has grayish color property “background”. The degree of separation
and it may be identified as smoke. This between pixels is identified by Euclidean
problem is rectified by Motion Property, Distance. Multi threshold algorithm is
where sky will be removed. based on channel wise approach,
reflectance of each channel value is used
Cheng (2011) [5] proposed a fire detection for threshold assumption and is applied to
system based on Neural Network; here each and every pixels of the image, smoke
neural network is used in detection pixels are marked and false pixels are
information for temperature, CO removed. Threshold value is set as 0.9 <=
concentration, and smoke density to channel 1 reflectance / channel 2
determine probability of three reflectance<= 1.5. Pixels which reach this
representative fire conditions. RBF neuron threshold are smoke pixels else are false
structure is used, the information regarding pixels and are removed.
temperature, CO concentration, and smoke
density are collected and data fusion is used (Refer Figure 6)
to generate fire signal decision. The
detectors have continuous analog outputs, Paulo Vinicius Koerich Borges [11]
when detection limit is exceeded the proposed a fire detection method based on
hardware circuit sends a local fire probabilistic method and classification.
indication to fusion center, this force the Computer vision based approach is used in
system detectors to generate final decision. this approach. Though this approach is used
Single-sensor detector is used to generate surveillance it is also used to automatic
the final decision. video classification for retrieval of fire
catastrophes in databases of newscast
(Refer Figure 4 & 5) content. There are large variations in fire
and background characteristics depending
Zhanqing (2001) [9] proposed another on the video instance. The proposed method
method using NN and Multi-threshold observes the frame-to-frame changes of
algorithm. In this method the NN not only low-level features describing potential fire
classify the smoke, sky, background but regions. These features include color, area
also generates a continuous random output size, surface coarseness, boundary
representing mixture of these. NN roughness, and skewness within estimated
consumes time in case of large areas so fire regions. Bayes classifier [12] is used
multi-threshold algorithm also used as well. for fire recognition. In addition, apriori
These two approaches may be combined or [12] knowledge of fire events captured in
used separately depending on the size of the videos is used to significantly improve the
area. Multilayer Perceptron Neural Network results. The fire region is usually located in

35
 ELK ASIA PACIFIC JOURNAL OF COMPUTER SCIENCE AND INFORMATION SYSTEMS

ISSN: 2454-3047 (Print) ISSN: 2394-0441 (Online) Volume 1 Issue 2 (2015)

……………………………………………………………………………………………………………………

the center of each frame. This fact is used

to model the probability of occurrence of [1] Kosmas Dimitropoulos, Panagiotis
fire. Barmpoutis, and Nikos Grammalidis
(2015). Spatio Temporal Flame
(Refer Figure 7) Modeling and Dynamic Texture
Analysis for automatic video-based fire
3. Conclusion detection, IEEE transactions on circuits
and systems for video technology, vol.
Different fire detection techniques have 25, no. 2.
been proposed for safety and protection of
the people and environment. It is very [2] Surya.T.S, Suchithra.M.S. Survey on
crucial to develop an appropriate detection Different Smoke Detection Techniques
system to avoid dangerous situation caused Using Image Processing. International
due to fire. Though fire detection using Journal of Research in Computer and
image produce satisfying result we now go Communication Technology, 16-19.
for fire detection to produce accurate result. [3] Turgay Celik, Huseyin Ozkaramanl, and
Wavelet based smoke detection is used for Hassan Demirel (2007). Fire and
smoke detection in video sequences of Smoke detection without Sensors:
outdoor environment. Covariance method is Image Processing based
for flame detection. This method use approach.15th European signal
temporally extended covariance matrices processing conference (eusipco 2007),
representing all the information together. Poznan, Poland, September 3-7.
The method works only well when the fire
is clearly visible. If the fire is small and if it [4] Osman Gunay, A. Enis C ， Etin,
is far away from the camera or covered by Yusuf Hakan, Habiboglu. Flame
dense smoke the method fails. Wavelet and Detection method in video using
Color model combined together and detect Covariance descriptors, IEEE
smoke earlier. Neural Network produces transactions, 1817-1820.
accurate result as it uses temperature,
smoke density and CO concentration. [5] CHENG Caixia, SUN Fuchun, ZHOU
Fuzzy based approach uses YCbCr and Xinquan (2011). One Fire Detection
HSV model and detects fire at an early Method Using Neural Networks,
stage. By these approaches we cannot Tsinghua Science and Technology,
completely protect the forest from fire but ISSN 1007-0214 05/17 31-35Volume
we reduce the level of damage. Perception 16, Number 1.
Neural Network along with Multi [6] Hamed Adab, Kasturi Devi Kanniah,
Threshold algorithm classified image pixels Karim Solaimani. Modeling forest fire
of cloud, land, smoke, and background and risk in the northeast of Iran using
produced accurate result of smoke. remote sensing and GIS techniques,
Springer Science+Business Media
References Dordrecht 2012.
36
 ELK ASIA PACIFIC JOURNAL OF COMPUTER SCIENCE AND INFORMATION SYSTEMS

ISSN: 2454-3047 (Print) ISSN: 2394-0441 (Online) Volume 1 Issue 2 (2015)

……………………………………………………………………………………………………………………

[7] Akshata Patil, Varsha Bhosale 92014). [11] Paulo Vinicius Koerich Borges (2010).
Survey of Local Binary Pattern for fire A Probabilistic Approach for Vision-
& smoke using Wavelet Based Fire Detection in Videos, IEEE
Decomposition. International Journal transactions on circuits and systems for
of Research in Engineering and video technology, vol. 20, no. 5.
Technology.
[12] Jiawei Han, Micheline Kamber, Jian
[8] Surapong Surit, Watchara Chatwiriya. Pei (2012). Data Mining Concepts and
Forest Fire Smoke Detection in Video Techniques, Third edition, 248-253,
Based on Digital Image Processing 350-351.
Approach.
[13] Yusuf Hakan Habibo˘glu, Osman
[9] Zhanqing Li, Alexandre Khananian, Günay, A. Enis Çetin (2011).
Robert H. Fraser, and Josef Cihlar. Covariance matrix-based fire and
Automatic Detection of Fire Smoke flame detection method in video”,
Using Artificial Neural Networks and Springer-Verlag, 17.
Threshold Approaches Applied to
AVHRR Imagery, IEEE transactions on [14] Rafael C. Gonzalez and Richard E.
geosciences and remote sensing, vol. Woods. Digital Image Processing.
39, no. 9, September 2001. Pearson publication, Third Edition.

[10] S. A. Christopher, M. Wang, T. A. [15] Prerna B. Pagar & A. N. Shaikh

Berendes, and R. M. Welch (1998). (2013). Real Time based Fire & Smoke
The 1985 biomass burning season in Detection without Sensor by Image
South America: Satellite remote Processing, International Journal of
sensing of fires, smoke, and regional Advanced Electrical and Electronics
radiative energy budgets, vol. 37, 661– Engineering (IJAEEE), 26-34.
678.

37
 ELK ASIA PACIFIC JOURNAL OF COMPUTER SCIENCE AND INFORMATION SYSTEMS

ISSN: 2454-3047 (Print) ISSN: 2394-0441 (Online) Volume 1 Issue 2 (2015)

……………………………………………………………………………………………………………………

List of Figures

Fig. 1. Detection of candidate fire region

Fig. 2. Smoke detection using LBP and Wavelet transform

38
 ELK ASIA PACIFIC JOURNAL OF COMPUTER SCIENCE AND INFORMATION SYSTEMS

ISSN: 2454-3047 (Print) ISSN: 2394-0441 (Online) Volume 1 Issue 2 (2015)

……………………………………………………………………………………………………………………

RGB image Y channel

Cb channel Cr channel

Fig. 3. YCbCr color model

Fig. 4. Multi-sensor information fusion detection system

39
 ELK ASIA PACIFIC JOURNAL OF COMPUTER SCIENCE AND INFORMATION SYSTEMS

ISSN: 2454-3047 (Print) ISSN: 2394-0441 (Online) Volume 1 Issue 2 (2015)

……………………………………………………………………………………………………………………

Fig. 5. RBF network structure [9]

Fig. 6. Multi threshold algorithm

40
 ELK ASIA PACIFIC JOURNAL OF COMPUTER SCIENCE AND INFORMATION SYSTEMS

ISSN: 2454-3047 (Print) ISSN: 2394-0441 (Online) Volume 1 Issue 2 (2015)

……………………………………………………………………………………………………………………

Fig. 7. Fire Detection Process

41

View publication stats