- Email: [email protected]
An enhanced real-time forest fire assessment algorithm based on video by using texture analysis
Accepted Manuscript Title: An Enhanced Real-time Forest Fire Assessment Algorithm Based On Video by Using Texture Analysis Author: Gudikandhula Narasimha Rao Peddada Jagadeeswara Rao Rajesh Duvvuru Sridhar Bendalam Roba Gemechu PII: DOI: Reference:
S2213-0209(16)30176-8 http://dx.doi.org/doi:10.1016/j.pisc.2016.06.037 PISC 322
To appear in: Received date: Accepted date:
13-6-2016 13-6-2016
Please cite this article as: Rao, G.N., Jagadeeswara Rao, P.,An Enhanced Realtime Forest Fire Assessment Algorithm Based On Video by Using Texture Analysis, Perspectives in Science (2016), http://dx.doi.org/10.1016/j.pisc.2016.06.037 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
An Enhanced Real-time Forest Fire Assessment Algorithm Based On Video by Using Texture Analysis Gudikandhula Narasimha Raoa, Peddada Jagadeeswara Raob, Rajesh Duvvuruc, Sridhar Bendalamd, Roba Gemechue Department of Geo-Engineering & Centre for Remote Sensing
a,b,c,d,e
Andhra University, Visakhapatnam, Andhra Pradesh, India.
ip t
a,b,c,d,e
Email:[email protected],[email protected],
cr
[email protected],[email protected], [email protected]
us
Abstract— As the human technology moved further, the risk of natural and man induced sudden damage increase exponentially. One of the most dangerous disasters is fire. In addition to its direct danger on human’s lives, fire consumes forests where trees that provide
an
humans with oxygen are destroyed. Every year, the large number of wildfires happening all over the world they burn forested lands, causing adverse ecological and social impacts. Early
M
warning and immediate responses are the only ways to avoid such type of disasters. This work describes a naïve method is used to detect flames in forest by using a Spatio Wildfire Prediction and Monitoring System (SWPMS). Basically, the fired information retrieving from
d
regions by using background subtraction and colour analysis. The fire behaviour is modelled
te
by texture analysis using computer vision systems. The Central Server should receives fired regions from the volunteer’s smart phone and use fired location coordinates, different angles
Ac ce p
of smart phone receives fired locations based on Google Earth API. Finally, kalman filter estimator computes the position vector of a moving object. Antennas or Satellite systems are grasping information from fire regions then GIS will be analysed those regions and send alert to local peoples of forest regions and NDRF team. Keywords- Flame Detection, Vital texture analysis, Remote Sensing, Geographic Information System, Semantic Web3.0.
1. Introduction
In this study, the problem of video based flame detection techniques has been deciphered. The important challenge in video-based flame detection describes chaotic and complex nature of the fire phenomenon. A dynamic approach is used to finds out reliability of the algorithm by using kalman filter. Firstly, different features retrieved from both dynamic texture analysis and Spatio-temporal flame modelling. Secondly, comparing between previous fired regions with current fires based on neighbouring blocks. Fire information
Page 1 of 5
will be transfer to Central system, that data will be classified by classification algorithm [1]. The final data reports of fire will be possessed on semantic web3.0 (Internet of Things). Geographic information system for quantitative evaluation of fired regions due to ground forest damaged information by using ArcGIS software. Fire information automatically sent
ip t
to nearest peoples, civilians and National Disaster Response Force (NDRF). The State of Andhra Pradesh is placed in the middle of eastern half of the Indian Peninsula lying between 120 41|–190 54| N latitudes and 760 46| -840 45| E longitudes.The Greater
cr
Visakhapatnam Municipal Corporation (GVMC) is one of the biggest fire region in Andhra Pradesh. The Kambalakonda Reserve Forest spread over 7,146 hectares located near
us
Visakhapatnam, Andhra Pradesh, India. In the year 2015, Kamabalakonda reserve forest has number of fire occurrences are there due to reason of environment conditions and man
an
made mistakes [2]. Generally, forest fires are mostly occurs from February to May. It has 8 times forest fires in February, 34 times March, 5 times in April, 5 times in May. Therefore
2. Fire Flame Detection Algorithm
M
based on temperature conditions most probable forest fire was done in the month of March.
The explosive things of forest fires are wood, leaves. Generally the flame is yellow to red,
d
sometimes it appears both the combination of yellow and red also with a sharp resolution due to the reason of dark brightness of forest. Moreover, flame is detecting from removal of back
te
ground disturbances that is removing of besides static background [3]. This section describes fire flame detections and static background subtraction techniques. This identifies object
Ac ce p
movements in forest field and removing background images those are non flame images in the field.
Algorithm:
Input: Location features La; a € A, Weather features Cb; b € B, First fire starting point at moment (a0; t0), maximum observation time is Tmax. Output: Maintain list of fired point regions according to moment of starting fire point with maintained array P = (a0; t0) Steps: Let k = 0; While P has points to simulate and the time of simulation does not exceed Tmax do for A1 neighbour of Ak do Let v = v (CAk, A1, tk) ; if v > 0 then
Page 2 of 5
Let Dt = d (Ak, A1) / v; if t +Dt < min{Tmax; t(A1)} then Change t (A1) ←t +Dt; Add (A1; t (A1)) to P;
ip t
Sort P according through time; end end
cr
end k = k+1
us
end
V is the fire spread rate. Da, a1 are distance between points and Ca, a1 are characteristics of
an
local and weather conditions. 2.1 Fire Colour Probability
The probability of each fire region, the non parametrically identified probabilities of each and
M
every pixel in field region. Estimated probabilities of each pixel in the block are used [4]. Moreover, total estimation of fire flame should be depending on the average probability estimation of each pixel and every pixel (i, j) in the region.
d
Q fire= (1/n) ∑i , j Q(i, j) pixel.
te
Where n is total pixels in the fire region and Q (i, j) is the fire colour probability of each
Ac ce p
2.2 Spatial Wavelet Analysis
Generally fire objects show good spatial resolution than natural fire coloured objects. Fire colour objects shows just images of either red nor yellow, In another case that is the combination of both red and yellow colours. So the demand of techniques should be placed at fire regions. Therefore we introduce edge detectors, interest points descriptors, etc. Here wavelet analysis using simple filters, those are achieved good computational efficiency because it can be implemented without any single multiplication that is normal register shifts. The spatial wavelet analysis describes energy at every pixel value is generated by low and high combinations [5]. Those are follows E (i, j ) = HL(i, j )2 + LH(i, j )2 + HH(i, j )2 For each block of fire region, the spatial wavelet analysis energy is calculated by using average energy resolutions of the pixels in the field [15]. Efire = (1/n) ∑i,j E(i,j)
Page 3 of 5
The background modelling is estimation of fire flames by using subtraction algorithm. The main theme of this one is to convert the video frame in to binary classification by using mat lab soft ware. Pre-processing is applying on forest fired field; it covers all pixels in static view as well as classifies the reaming fired regions based on object tracking or flame
ip t
detection. 3. Results and Discussions
SWPMS identifies fired image, stores location coordinates of fire, rotation angles of smart
cr
phone and location timings. The basic knowledge of fired region is depending upon the area latitude and longitude. Therefore, system must be careful with that coordinates and finally
us
conclude the fired regions. Secondly, the SWPMS describes list out types of vegetation in
Ac ce p
(a)
te
d
M
an
forest and identifies which vegetation is caused to more fires.
(b)
(c)
Figure 1. (a) Identify flame analysis by using coloured video in Kailasapuram and Simhachalam hills, (b) Object tracking, (c) Identify flame analysis by using gray scale video. The Andhra Pradesh northern parts will be biggest threat due to burn agriculture. The above figure 1 shows Identify flame analysis by using coloured video in Kailasapuram and Simhachalam hills in Visakhapatnam. Where Visakhapatnam has 23.3% forest burnt area. That categorization refers patch analysis of fired lands information; those are all less than 25 ha. The burnt land is greater than 400 shows 5545.3 km2. In Kalman filter, spatial texture conservation of fire regions are divided in to three clusters including the RGB,YCbCr, HSV and so on. Here the fire region will be converted to other colour spaces, where RGB is the basic colour. The fire region of HSV describes brightness of the fire [6]. The figure 2 shows Visakhapatnam is most forest fire prone area in Andhra Pradesh.
Page 4 of 5
ip t
cr
Figure 2. District wise forest burnt area statistics of coastal Andhra Pradesh. 4. Conclusion
us
A novel method is proposed for flame detection based on the combination of features extracted from spatiotemporal flame modelling and dynamic texture analysis. Reduces the computational
an
cost required for the dynamic texture analysis through redundant data reduction. Determines the time of the fire incident, localizes the exact position of the fire in the image. The flame behaviour modelling is introduced to identify the colour of the fire, the motion characteristics of
M
flame as well as the random variations of its appearance. Fire information will be transfer to Central system, that data will be classified by classification algorithm. The final data reports of
d
fire will be possessed on semantic web 3.0. Geographic information system for quantitative evaluation of forest quarter fire danger based on ground forest taxation data, with use of
te
specialized software ArcGIS. Fire information automatically sent to nearest peoples and National Disaster Response Force (NDRF).
Ac ce p
References
[1] Ananthakrishnan, T.N. (Ed.), 2001. Insects and Plant Defence Dynamics. Science Publishers Inc.,Enfield, NH, USA.
[2] Gudikandhula Narasimha Rao, et.al “A Drone Remote Sensing for Virtual Reality Simulation System for Forest Fires: Semantic Neural Network Approach”
IOP Conference
Series: Materials Science and Engineering, ISSN 157-899X, 2016. [3] A. Elgammal, D. Harwood, and L. Davis, “Non-parametric model for background subtraction,” in Proc. 6th Eur. Conf. Comput. Vis., Dublin, Ireland, Jun. 2000, pp. 751–767. [4] C.B. Liu and N. Ahuja, “Vision based fire detection,” in Proc. ICPR, Aug. 2004. [5] G. Doretto, et.al “Dynamic textures,” Int. J. Comput. Vis., vol. 51, pp. 91–109, Feb. 2003. [6] Rao, Gudikandhula Narasimha, and P. Jagdeeswar Rao. "A Clustering Analysis for Heart Failure Alert System Using RFID and GPS." ICT and Critical Infrastructure: Proceedings of the 48th Annual Convention of Computer Society of India-Vol I. Springer.
Page 5 of 5
S2213-0209(16)30176-8 http://dx.doi.org/doi:10.1016/j.pisc.2016.06.037 PISC 322
To appear in: Received date: Accepted date:
13-6-2016 13-6-2016
Please cite this article as: Rao, G.N., Jagadeeswara Rao, P.,An Enhanced Realtime Forest Fire Assessment Algorithm Based On Video by Using Texture Analysis, Perspectives in Science (2016), http://dx.doi.org/10.1016/j.pisc.2016.06.037 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
An Enhanced Real-time Forest Fire Assessment Algorithm Based On Video by Using Texture Analysis Gudikandhula Narasimha Raoa, Peddada Jagadeeswara Raob, Rajesh Duvvuruc, Sridhar Bendalamd, Roba Gemechue Department of Geo-Engineering & Centre for Remote Sensing
a,b,c,d,e
Andhra University, Visakhapatnam, Andhra Pradesh, India.
ip t
a,b,c,d,e
Email:[email protected],[email protected],
cr
[email protected],[email protected], [email protected]
us
Abstract— As the human technology moved further, the risk of natural and man induced sudden damage increase exponentially. One of the most dangerous disasters is fire. In addition to its direct danger on human’s lives, fire consumes forests where trees that provide
an
humans with oxygen are destroyed. Every year, the large number of wildfires happening all over the world they burn forested lands, causing adverse ecological and social impacts. Early
M
warning and immediate responses are the only ways to avoid such type of disasters. This work describes a naïve method is used to detect flames in forest by using a Spatio Wildfire Prediction and Monitoring System (SWPMS). Basically, the fired information retrieving from
d
regions by using background subtraction and colour analysis. The fire behaviour is modelled
te
by texture analysis using computer vision systems. The Central Server should receives fired regions from the volunteer’s smart phone and use fired location coordinates, different angles
Ac ce p
of smart phone receives fired locations based on Google Earth API. Finally, kalman filter estimator computes the position vector of a moving object. Antennas or Satellite systems are grasping information from fire regions then GIS will be analysed those regions and send alert to local peoples of forest regions and NDRF team. Keywords- Flame Detection, Vital texture analysis, Remote Sensing, Geographic Information System, Semantic Web3.0.
1. Introduction
In this study, the problem of video based flame detection techniques has been deciphered. The important challenge in video-based flame detection describes chaotic and complex nature of the fire phenomenon. A dynamic approach is used to finds out reliability of the algorithm by using kalman filter. Firstly, different features retrieved from both dynamic texture analysis and Spatio-temporal flame modelling. Secondly, comparing between previous fired regions with current fires based on neighbouring blocks. Fire information
Page 1 of 5
will be transfer to Central system, that data will be classified by classification algorithm [1]. The final data reports of fire will be possessed on semantic web3.0 (Internet of Things). Geographic information system for quantitative evaluation of fired regions due to ground forest damaged information by using ArcGIS software. Fire information automatically sent
ip t
to nearest peoples, civilians and National Disaster Response Force (NDRF). The State of Andhra Pradesh is placed in the middle of eastern half of the Indian Peninsula lying between 120 41|–190 54| N latitudes and 760 46| -840 45| E longitudes.The Greater
cr
Visakhapatnam Municipal Corporation (GVMC) is one of the biggest fire region in Andhra Pradesh. The Kambalakonda Reserve Forest spread over 7,146 hectares located near
us
Visakhapatnam, Andhra Pradesh, India. In the year 2015, Kamabalakonda reserve forest has number of fire occurrences are there due to reason of environment conditions and man
an
made mistakes [2]. Generally, forest fires are mostly occurs from February to May. It has 8 times forest fires in February, 34 times March, 5 times in April, 5 times in May. Therefore
2. Fire Flame Detection Algorithm
M
based on temperature conditions most probable forest fire was done in the month of March.
The explosive things of forest fires are wood, leaves. Generally the flame is yellow to red,
d
sometimes it appears both the combination of yellow and red also with a sharp resolution due to the reason of dark brightness of forest. Moreover, flame is detecting from removal of back
te
ground disturbances that is removing of besides static background [3]. This section describes fire flame detections and static background subtraction techniques. This identifies object
Ac ce p
movements in forest field and removing background images those are non flame images in the field.
Algorithm:
Input: Location features La; a € A, Weather features Cb; b € B, First fire starting point at moment (a0; t0), maximum observation time is Tmax. Output: Maintain list of fired point regions according to moment of starting fire point with maintained array P = (a0; t0) Steps: Let k = 0; While P has points to simulate and the time of simulation does not exceed Tmax do for A1 neighbour of Ak do Let v = v (CAk, A1, tk) ; if v > 0 then
Page 2 of 5
Let Dt = d (Ak, A1) / v; if t +Dt < min{Tmax; t(A1)} then Change t (A1) ←t +Dt; Add (A1; t (A1)) to P;
ip t
Sort P according through time; end end
cr
end k = k+1
us
end
V is the fire spread rate. Da, a1 are distance between points and Ca, a1 are characteristics of
an
local and weather conditions. 2.1 Fire Colour Probability
The probability of each fire region, the non parametrically identified probabilities of each and
M
every pixel in field region. Estimated probabilities of each pixel in the block are used [4]. Moreover, total estimation of fire flame should be depending on the average probability estimation of each pixel and every pixel (i, j) in the region.
d
Q fire= (1/n) ∑i , j Q(i, j) pixel.
te
Where n is total pixels in the fire region and Q (i, j) is the fire colour probability of each
Ac ce p
2.2 Spatial Wavelet Analysis
Generally fire objects show good spatial resolution than natural fire coloured objects. Fire colour objects shows just images of either red nor yellow, In another case that is the combination of both red and yellow colours. So the demand of techniques should be placed at fire regions. Therefore we introduce edge detectors, interest points descriptors, etc. Here wavelet analysis using simple filters, those are achieved good computational efficiency because it can be implemented without any single multiplication that is normal register shifts. The spatial wavelet analysis describes energy at every pixel value is generated by low and high combinations [5]. Those are follows E (i, j ) = HL(i, j )2 + LH(i, j )2 + HH(i, j )2 For each block of fire region, the spatial wavelet analysis energy is calculated by using average energy resolutions of the pixels in the field [15]. Efire = (1/n) ∑i,j E(i,j)
Page 3 of 5
The background modelling is estimation of fire flames by using subtraction algorithm. The main theme of this one is to convert the video frame in to binary classification by using mat lab soft ware. Pre-processing is applying on forest fired field; it covers all pixels in static view as well as classifies the reaming fired regions based on object tracking or flame
ip t
detection. 3. Results and Discussions
SWPMS identifies fired image, stores location coordinates of fire, rotation angles of smart
cr
phone and location timings. The basic knowledge of fired region is depending upon the area latitude and longitude. Therefore, system must be careful with that coordinates and finally
us
conclude the fired regions. Secondly, the SWPMS describes list out types of vegetation in
Ac ce p
(a)
te
d
M
an
forest and identifies which vegetation is caused to more fires.
(b)
(c)
Figure 1. (a) Identify flame analysis by using coloured video in Kailasapuram and Simhachalam hills, (b) Object tracking, (c) Identify flame analysis by using gray scale video. The Andhra Pradesh northern parts will be biggest threat due to burn agriculture. The above figure 1 shows Identify flame analysis by using coloured video in Kailasapuram and Simhachalam hills in Visakhapatnam. Where Visakhapatnam has 23.3% forest burnt area. That categorization refers patch analysis of fired lands information; those are all less than 25 ha. The burnt land is greater than 400 shows 5545.3 km2. In Kalman filter, spatial texture conservation of fire regions are divided in to three clusters including the RGB,YCbCr, HSV and so on. Here the fire region will be converted to other colour spaces, where RGB is the basic colour. The fire region of HSV describes brightness of the fire [6]. The figure 2 shows Visakhapatnam is most forest fire prone area in Andhra Pradesh.
Page 4 of 5
ip t
cr
Figure 2. District wise forest burnt area statistics of coastal Andhra Pradesh. 4. Conclusion
us
A novel method is proposed for flame detection based on the combination of features extracted from spatiotemporal flame modelling and dynamic texture analysis. Reduces the computational
an
cost required for the dynamic texture analysis through redundant data reduction. Determines the time of the fire incident, localizes the exact position of the fire in the image. The flame behaviour modelling is introduced to identify the colour of the fire, the motion characteristics of
M
flame as well as the random variations of its appearance. Fire information will be transfer to Central system, that data will be classified by classification algorithm. The final data reports of
d
fire will be possessed on semantic web 3.0. Geographic information system for quantitative evaluation of forest quarter fire danger based on ground forest taxation data, with use of
te
specialized software ArcGIS. Fire information automatically sent to nearest peoples and National Disaster Response Force (NDRF).
Ac ce p
References
[1] Ananthakrishnan, T.N. (Ed.), 2001. Insects and Plant Defence Dynamics. Science Publishers Inc.,Enfield, NH, USA.
[2] Gudikandhula Narasimha Rao, et.al “A Drone Remote Sensing for Virtual Reality Simulation System for Forest Fires: Semantic Neural Network Approach”
IOP Conference
Series: Materials Science and Engineering, ISSN 157-899X, 2016. [3] A. Elgammal, D. Harwood, and L. Davis, “Non-parametric model for background subtraction,” in Proc. 6th Eur. Conf. Comput. Vis., Dublin, Ireland, Jun. 2000, pp. 751–767. [4] C.B. Liu and N. Ahuja, “Vision based fire detection,” in Proc. ICPR, Aug. 2004. [5] G. Doretto, et.al “Dynamic textures,” Int. J. Comput. Vis., vol. 51, pp. 91–109, Feb. 2003. [6] Rao, Gudikandhula Narasimha, and P. Jagdeeswar Rao. "A Clustering Analysis for Heart Failure Alert System Using RFID and GPS." ICT and Critical Infrastructure: Proceedings of the 48th Annual Convention of Computer Society of India-Vol I. Springer.
Page 5 of 5