Sie sind auf Seite 1von 4

Survey on Different Smoke Detection Techniques Using Image

Processing
Surya.T.S, Suchithra.M.S
P.G. Student, Assistant Professor
Department of Computer Science & Engineering
Christ Knowledge City , Ernakulam, India
surya.tharayil@gmail.com, suchithrams194@gmail.com

different characteristics of smoke. Smoke detection


Abstract- The most significant parts of protective method is one of the efficient and challenging
and monitoring systems are the fire detection problems mainly in open environment, such as port,
systems. Fire detection is very important for the chemical plants and power plants as they cause harm
safety of the people. The main causes of disasters are to surrounding areas.
the failure in fire detection and lack of information.
Smoke is the good indicator of fire. Fire can be In such areas, fire or flame detection methods [2] are
detected effectively by using different image limited because fire or flames sources cannot always
processing techniques. In image processing, images fall properly into the range of sensors due to their
like video frames or photographs are the inputs and positions and sizes. So smoke detection is good and
output may be either an image or image effective for early fire detection in such cases. In
characteristics. Various tasks like classification, conventional methods heat or smoke sensors are used
extracting the features, recognizing different patterns for the detection of fire. These sensors had some
can be performed using image processing. Image limitations like time and they cannot cover a wide
processing techniques greatly help to detect fire in a range of area. To overcome such limitations video
good manner and thereby avoid dangerous situation fire detection systems are used.
according to the output. This paper describes
different techniques used for fire and smoke Different image processing can be used to detect
detection. smoke. In image processing image is taken as input
and the output may either an image or parameters or
Index Terms- Wavelet Transform, Support Vector characteristics of an image. It is the processing of 2D
Machine, Covariance Descriptors, Multi-Layers image. Various tasks like classification, extracting
Perceptron, Image Fusion the features, recognizing different patterns can be
performed using image processing. The features and
I. INTRODUCTION textures of smoke can extract using image processing
techniques. By these techniques the dangerous
Fire detection is very crucial for the safety of the situations caused due to fire can be avoided and save
people. Fire detection can be possible through the people. This paper describes different smoke
different image processing techniques. Smoke is the detection techniques and their comparison.
good indicator of fire. By analyzing the features and
other characteristics present in the image, fire and II. LITERATURE REVIEW
smoke can be detected. This paper describes different
fire detection techniques, its advantages and R. Gonzalez-Gonzalez et al. [6] proposed a method to
disadvantages. Smoke is a significant hint for early detect fire by smoke detection based on wavelet. In
fire detection. Smoke become visible before flames this smoke detection method, image processing on
on forest fires in most of the times, and can detect video signals is proposed. The SWT transform is
fires in their early stages to reduce dangerous used for the area detection of ROI’s. This method
situation caused by fire. comprises of three steps.

Detection methods based on image are more effective In the first step, preprocessing is performed and the
than other techniques which use certain sensor image is resized and transformed to grayscale image.
devices. Characteristics of smoke need to be Finally indexed the image using indexation. The
considering when dealing with image information. second step involves high frequencies of an image is
Semi–transparency, response to environmental eliminated using SWT and reconstruct the image by
conditions, unsteady nature of smoke shape are the inverse SWT. Inorder to group the intensity colors

All copy rights Reserved by NATCOMM - 2014, Christ Knowledge City , Mannoor, India.
Published by IJRCCT (www.ijrcct.org) Page 16
that are closed to each other is the main purpose of detection of flames. Background subtraction method
image indexation. is not used in this approach.

Histogram analysis is used to determine the To detect fire, divides the video into spatio temporal
indexation levels. After that compare the image with blocks and covariance features are computed from
a non-smoke frame and selecting those pixels that are these blocks. Using an SVM classifier, the flame
change from one scene to another. The final stage colored region are classified by using the spatial and
consists of smoke verification algorithm inorder to temporal characteristics. These classified flame
determine whether ROI is increasing its area and to colored regions are tested using video data that
reduce the generation of false alarm. These three contain flames and flame colored objects.
steps are combined together to form the final result.
Hidenori Maruta et al. [7] proposed another method For the classification of pixel colors chromatic color
for smoke detection based on support vector model [4] is used and analysed fire colored pixels.
machine. In this approach robust and novel smoke Object detection and texture classification [5] are
detection method is proposed using support vector performed by applied covariance descriptors [5].
machine. Firstly preprocessing is performed by Inorder to define spatio-temporal blocks temporal
extracting moving objects of images. The derivatives are calculated along with spatial
preprocessing consists of five steps: image parameters. Spatial temporal blocks can be defined
subtraction and using covariance matrices. Then compute the
covariance values of the pixel property vectors in
accumulation, image binarization, morphological spatio-temporal blocks.
operation, extraction of Feret’s regions and creation
of the image mask. Image subtraction is used to For classification, support vector machine is trained.
extract regions of moving object. Inorder to eliminate According to the number positively classified video
noise like regions binarization and morphological blocks and their positions, confidence value is
operations are used. The position and approximated determined for fire detection. This method is
shape of the object is obtained by identifying Feret’s computationally efficient. Covariance approach is
diameter is called feret’s region. well suited for detection of flames. Drawback of this
method is that, it is well suited when the fire is
clearly visible. If the fire is far away from the camera
After preprocessing, perform texture analysis and and covered with dense smoke, this method performs
extract the texture features. These texture features poorly.
become the component of feature vector. Feature
vector is applied as the input vector and support Pietro Morerio et al. [10] proposed another method
vector machine is used to classify smoke or not. that uses color features and motion analysis to detect
Smoke detection method involves three steps: fire and smoke. Smoke is an good indicator of fire
analysing texture features, discrimination of Feret’s and can easily determine the presence of fire. In this
region using support vector machine and time method classify the pixel as fire, smoke or
accumulation. background pixel and then clustered the pixels.

Inorder to extract feature vectors of the image,


texture analysis is performed in this method. Support Then analyze the internal motion of region to identify
vector machine is used to classify smoke or non- whether they are dangerous. In this approach a fire
smoke from the extracted image. The main advantage and smoke detection system is proposed and contains
of this method is that more accurate extraction of five modules:
smoke areas in image can be obtained using SVM.
change detection, fire features extraction, smoke
Y. Habiboglu et al. [9] proposed another method that features extraction and chaotic motion detection.
uses covariance descriptors for fire detection. In this
method, color, spatial and domain information are
combined by using covariance descriptors for each
spatio-temporal block. The blocks are generated by
dividing the flame colored region into 3D regions.
This method used a covariance matrix for the

All copy rights Reserved by NATCOMM - 2014, Christ Knowledge City , Mannoor, India.
Published by IJRCCT (www.ijrcct.org) Page 17
detection of short range smoke visible and IR image
fusion algorithm is used. Scope of this paper is to
detect the smoke as an indicator of fire. Here visible
and infrared images are combined togeth-er to
distinguish smoke from smoke like objects. Earlier
ap-proach that uses sensor or visible images only
gives false alarms.

Visible images capture both smoke and smoke like


ob-jects. Infrared images do not capture smoke.
Integrating these images give correct information
about smoke. Objective of this paper is to save
people, forest from the fire. By this meth-od
generation of false alarm can be reduced to a great
Fig. 1. Fire and smoke detection module’s structure extent.
In change and motion detection module background
subtraction is used to identify pixels which behave The proposed algorithm consists of two phases. In the
differently from the normal behavior of the first phase combine visual and thermal information of
background image. In color spaces and features the smoke and potential smoke mask is generated.
extraction module performs a separation between fire PSM is again analyzed to differentiate true and false
pixel and smoke pixels from other pixels. In the case alarms. This method is very efficient and detects
of smoke, separation is performed by swapping to smoke successfully. Improves the fire detection rate
L*a*b space [8]. Separation of fire pixels is and reduces the generation of false alarm. The
accomplished by moving to YCbCr. Then pixels drawback of this method is Correct and punctual
separated by change detection and feature extraction detec-tion of fire is not possible and comparison is
are joined together to get connected regions in region required to identify smoke.
growing module. After that chaotic motion analysis is
performed.

After this analysis data fusion is conducted in which


a pre-alarm is generated. Main purpose of data fusion
is to filter false alarms. Data fusion task is
accomplished by classifying the pre-alarm rectangles
using a multi-layered perceptron (MLP) [1].MLP
consists of multi-layered network includes input
layer, hidden layer and output layers. In this method
three output neurons are considered and it represents
three pre-alarm states: pre-alarm of smoke type, pre-
alarm of fire type and no pre-alarm. Finally, a simple
tracking algorithm is used on superposition of pre-
alarm regions [3] to filter false alarms.

MehdiTorabnezhad et al. [11] proposed another


method that used image fusion technique to detect Fig.2. Flow chart of Smoke detection
smoke. In this method, combine visual and thermal algorithm
information to improve the rate of fire detection. The
invisibility of smoke in LWIR image can distinguish III. CONCLUSION
smoke from smoke like objects. In-frared images do
not detect smoke in the images but can detect smoke Different smoke detection techniques have been
like object. By combining visible and IR images proposed for safety and protection of the people and
smoke can be distinguished. Based on characteristics environment. It is very crucial to develop an
of visual and thermal smoke images a potential appropriate detection system to avoid dangerous
smoke mask is created. In-order to reduce false situation caused due to fire. Wavelet based smoke
alarms, PSM is further analyzed by dis-order detection approach is useful for smoke detection in
measurements and energy calculations. For the outdoor video sequences. This approach is robust and

All copy rights Reserved by NATCOMM - 2014, Christ Knowledge City , Mannoor, India.
Published by IJRCCT (www.ijrcct.org) Page 18
reduces the gen-eration of false alarms due to the classification,” Computer Vision– ECCV 2006, pp.
smoke verification al-gorithm. Smoke detection 589–600, 2006.
method using support vector ma-chine is based on the [6] R. Gonzalez-Gonzalez, V. Alarcon-Aquino,
texture analysis and the support vector classifier is R. Rosas-Romero, O. Starostenko, J. Rodriguez-
also efficient method. In this method, the texture Asomoza, “Wavelet-Based Smoke Detection in
features are used as the input vector and determine Outdoor Video Sequences,” 978-1-4244-7773-
whether the candidate regions are smoke regions or 9/10/$26.00 ©2010 IEEE
not with non–lin-ear SVM. Here classifiers are used
to classify smoke and other objects. Covariance [7] Hidenori Maruta, Akihiro Nakamura, Fujio
method is mainly suited for flame detec-tion because Kurokawa, “A Novel Smoke Detection Method
flames exhibit random behavior. In this method use Using Support Vector Machine,” 978-1-4244-6890-
temporally extended covariance matrices 4/10/$26.00 ©2010 IEEE.
representing all the information together. The method
works well when the fire is clearly visible and in [8] CELIK Turgay, “Fast and efficient method
close range. Background subtraction method is not for fire detection using image processing,” ETRI
used in this approach. If the fire is small and far away journal, vol. 32, no. 6, pp. 881– 890, 2010, eng.
from the camera or covered by dense smoke the [9] Y. Habiboglu, O. Gunay, and A. Cetin,
method perform poorly. This method is "Flame detection method in video using covariance
computationally efficient. Fire and smoke detection descriptors," IEEE International Conference on
method based on color features and mo-tion analysis Speech and Signal Processing, pp. 1817-1820, 2011.
classify the pixels as fire, smoke or background
pixels. In this method, information of visible and [10] Pietro Morerio, Lucio Marcenaro, Carlo S.
infrared im-ages is combined for the smoke detection Regazzoni, Gianluca Gera, "Early fire and smoke
in short range. Smoke and smoke like objects can be detection based on colour features and motion
distinguished by comb-ing visible and infrared Analysis," 978-1-4673-2533-2/12/$26.00 ©2012
images. This method is very efficient and detects IEEE.
smoke successfully. Improves the fire detection rate [11] MehdiTorabnezhad, Ali Aghagolzadeh,
and reduces the generation of false alarm. The HadiSeyedarabi, “Visible and IR image fusion
drawback of this method is Correct and punctual algorithm for short range smoke detection,” IEEE
detection of fire is not possible and comparison is International Conference on Robotics and
required to identify smoke. Mechatronics, 2013.

REFERENCES

[1] Fa-Long Luo and Rolf Unbehauen, Applied


Neural Networks For Signal Processing, Cambridge
University Press, New York, NY, USA, 1999.

[2] W. Phillips III, M. Shah, N. V. Lobo:”Flame


recognition in video”, Pattern Recognition Lett. 23,
pp. 319–327, 2002.
[3] L. Marchesotti, L. Marcenaro, and C. S.
Regazzoni, “Tracking and counting multiple
interacting pedestrian in indoor scenes,” in 3rd IEEE
International Workshop on Performance Evaluation
of Tracking and Surveillance, PETS02, Copenhagen,
Denmark, June 2002.

[4] Thou-Ho Chen, Ping-HsuehWu, and Yung-


Chuen Chiou,“An early fire-detection method based
on image processing,” in ICIP ’04., 24-27 2004, vol.
3, pp. 1707 – 1710 Vol. 3.

[5] O. Tuzel, F. Porikli, and P. Meer, “Region


covariance: A fast descriptor for detection and

All copy rights Reserved by NATCOMM - 2014, Christ Knowledge City , Mannoor, India.
Published by IJRCCT (www.ijrcct.org) Page 19

Das könnte Ihnen auch gefallen