Research ArticleForest Fire Detection Using a Rule-Based Image ProcessingAlgorithm and Temporal Variation
Mubarak A I Mahmoud 1 and Honge Ren 12
1College of Information and Computer Engineering Northeast Forestry University Harbin Heilongjiang 150040 China2Forestry Intelligent Equipment Engineering Research Center Harbin Heilongjiang 150040 China
Correspondence should be addressed to Honge Ren nefu rhe163com
Received 26 April 2018 Revised 18 September 2018 Accepted 26 September 2018 Published 21 October 2018
Academic Editor Vittorio Bianco
Copyright copy 2018 MubarakA I Mahmoud andHonge RenThis is an open access article distributed under the Creative CommonsAttribution License which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited
Forest fires represent a real threat to human lives ecological systems and infrastructure Many commercial fire detection sensorsystems exist but all of them are difficult to apply at large open spaces like forests because of their response delay necessarymaintenance needed high cost and other problems In this paper a forest fire detection algorithm is proposed and it consistsof the following stages Firstly background subtraction is applied to movement containing region detection Secondly convertingthe segmented moving regions from RGB to YCbCr color space and applying five fire detection rules for separating candidatefire pixels were undertaken Finally temporal variation is then employed to differentiate between fire and fire-color objects Theproposed method is tested using data set consisting of 6 videos collected from Internet The final results show that the proposedmethod achieves up to 9663 of true detection ratesThese results indicate that the proposed method is accurate and can be usedin automatic forest fire-alarm systems
1 Introduction
Forest fire detection systems are gaining a lot of attentionbecause of the continual threat from fire to both economicproperties and public safety [1] Hundreds of millions ofhectares are destroyed by wildfires each year [2] and over200000 forest fires happen every year in the world Forestfires destroy a total area of 35 to 45 million km2 [3] Increasein forest fires in forest areas around the world has resulted inan increased motivation for developing fire warning systemsfor the early detection of wildfires [1] Sensor technologyhas been widely used in fire detection usually dependingon sensing physical parameters such as changes in pressurehumidity and temperature as well as chemical parameterssuch as carbon dioxide carbon monoxide and nitrogendioxide However it is hard to apply these systems in largeopen areas for a variety of reasons such as high cost energyusage by the sensors and the necessary proximity of thesensor to the fire for accurate sensing resulting in physicaldamage to the sensors [4] In addition sensor methods have ahigh false alarms rate and their response time is quite big [5]
There are numerous motivating factors for the use of animage processing based method of fire detection The firstfactor is the rapid development of digital camera technologyand CCD or CMOS digital cameras which has resulted ina rapid increase in image quality and decreased cost of thecameras The second factor is that digital cameras can coverlarge areas with excellent results Third the response time ofimage processing models is better than that of existing sensormodels Finally the overall cost of image processing systemsis lower than existing systems
11 Related Studies Several fire detection algorithms havebeen proposed by various researchers Thou-Ho et al [6]presented fire detection algorithm which combines thesaturation channel of the HSV color and the RGB color Thisalgorithm employs three rules (RgeGgtB) (RgeRT) and (Sge((255-R) lowastSTRT) Determination of the two thresholds RTand ST is required The certain values range is from 115 to 135for RT and from 55 to 65 for ST based onmany investigationalresults done by the authors This method is computationallysimple compared to the other algorithms however it suffers
HindawiMathematical Problems in EngineeringVolume 2018 Article ID 7612487 8 pageshttpsdoiorg10115520187612487
2 Mathematical Problems in Engineering
Get image sequence
Apply background-subtraction Update background model
Moving Obj available
RGB to YCbCr convert
Apply fire detection rules
fire detected Set fire alarm
NO
NO
YES
YES
temporal variation
Figure 1 The proposed fire detection method
from false-positive alarms in case of moving fire-like objectsDios et al [2] presented an optical model used to detect forestfires and measure the properties of the fire such as flameheight fire front fire base width and flame inclination angleThis system is very good nevertheless it is very expensivebecause it consists of infrared cameras and other technologiessuch as GPS and telemetry sensors Yinglian et al [7]proposed forest fire disaster prevention algorithm based onimage processing This algorithm depends on fire and smokecolor properties to identify fire Yinglianrsquos algorithm is goodbut the smoke spreads quickly and it hasmany different colorswhich depend on the burning material thus the false alarmrate rises
In this paper a forest fire detection algorithm is proposedThe algorithm uses YCbCr color space since it effectively sep-arates luminance from chrominance and is able to separatehigh temperature fire center pixels because the fire at the hightemperature center region is white The final results showthat the proposed system has good detection rates and fewerfalse alarms which are the main crucial problems of the mostexisting algorithms
This paper is organized as follows Section 2 describes theproposed fire detection method Section 3 presents the resultsand computational complexity of the proposed algorithm
and Section 4 summarizes the work that has been carried outin this study and potential future direction
2 Materials and Methods
This section presents the proposed forest fire detectionalgorithm It consists of the following main stages the firststep receives the input video from the input device the secondstep applies movement containing region detection based onbackground subtraction (MRDB) the third step converts theinput image sequence from RGB to YCbCr color space andthe fourth step applies the fire detection rules and temporalvariation A fire alarm is activated if all detection conditionsare satisfied The proposed algorithm stages will be describedin detail Figure 1 shows the proposed algorithm flowchart
21 Movement Containing Region Detection Based on Back-ground Subtraction (MRDB) Detecting moving regions isa key factor in most of the video based fire detectionsystems because fire boundaries continuously fluctuate Sobackground subtraction is used to select candidate regions offire A pixel located at (x y) is supposed to be moving if thefollowing condition is satisfied [8]
1003816100381610038161003816119868119899 (119909 119910) minus 119861119899 (119909 119910)1003816100381610038161003816 gt 119905ℎ119903 (1)
Mathematical Problems in Engineering 3
(a) (b)
Figure 2 Example of MRDB (a) Frame before MRDB (b) frame after applying MRDB
(a) (Y) (Cb) (Cr)
Figure 3 (a) An RGB image and its (Y) (Cb) and (Cr) channels
119868119899(x y) represents the intensity value of the pixel at location(x y) in the nth gray-level for the current frame and Bn (xy) represent the background intensity value at the same pixellocation and thr is a threshold value experimentally set to 3The background is continuously updated using (2)
where 119861119899+1(x y) and 119861119899(x y) represent the intensity of thepixel value at location (x y) for the current and previousbackgrounds Figure 2 shows an example of MRDB
22 Converting RGB Images to YCbCr Due to the fact thatdifferent kinds of moving objects can be included after apply-ing background subtraction such as trees animals birds andpeople therefore images from the background subtractionstage are converted to YCbCr [9] to select candidate fireregions using (3) Figure 3 shows original RGB image (a) andYCbCr component The mean values of the YCbCr channelare then calculated using (4) (5) and (6)
where 119884119898119890119886119899 119862119887119898119890119886119899 and 119862119903119898119890119886119899 are the mean values for theYCbCr channels Y (x y) Cb (x y) and Cr (x y) are YCbCrchannel values for pixel at specific location (x y) and NlowastMis the total number of pixels
23 Fire Color Pixel Detection Rules In any fire image pixelsthe red color value is larger than green and green is largerthan blue as illustrated in Figure 4 (a) is a fire image and(b) is the RGB channels histogram for the same image Thisfact is represented in RGB color space as RgtGgtB and can beconverted to YCbCr using the flowing equations
Also the Y component value is greater than the mean Ycomponent of the same image and the Cb component issmaller than the mean Cb of same image while the Cr
4 Mathematical Problems in Engineering
(a)
0 50 100 150 200 250 3000
200
400
600
800
1000
1200
1400
1600
1800
Blue channelGreen channelRed channel
(b)
Figure 4 (a) RGB fire image and (b) the Histogram of the same image channels
component is greater than the mean Cr mean componentThis fact can be represented by the flow
where F(x y) can be any pixel on the image Y119898 Cb119898 andCr119898 are the mean values for Y Cb and Cr respectively
The Cb component as shown is predominantly ldquoblackrdquoand the Cr component is ldquowhiterdquo This idea can be repre-sented by the following equations
119865 (119909 119910) = 1 if 1003816100381610038161003816119862119887 (119909 119910) minus 119862119903 (119909 119910)1003816100381610038161003816 ge 1205910 119874119905ℎ119890119903119908119894119904119890 (10)
119865 (119909 119910)
= 1 (119862119887 (119909 119910) le 120)⋂ (119862119903 (119909 119910) ge 150)0 119874119905ℎ119890119903119908119894119904119890
(11)
where 120591 is a constant specified in [10] using receiver oper-ating characteristic (ROC) by applying different values of 120591in the range [1 100] To measure the ldquotrue detection raterdquoand ldquofalse detectionrdquo data sets consisting of 500 images (300of them being images of a forest fire 200 nonfire images)collected from Internet were used Only (10) was used withdifferent values of 120591 in the range [1 100] to detect binaryimages of the candidate fire region 120591 was selected as 120591 = 70resulting in a true detection rate of more than 90 and falsedetection of less than 40
Figure 5 shows the application of detection rules (7)through (11)
24 Temporal Variation Using color models alone is notenough to identify fire correctly because there are severalobjects that share the samefire color such as red leaves desertand other red moving objects The main difference betweenactual fire and the fire-color objects is the nature of theirmotion Shape and size of the flame are totally changeablebecause of burning materials and airflow thus it produceshigher temporal variation In contrast rigid bodiesrsquo motionproduces lower temporal variation Therefore it is possibleto differentiate between the fire pixels and the fire color Todetect a fire movement the difference between successiveframes was analyzed Suppose a video sequence consisting ofn frames and the average temporal variation is defined as [11]
Δ (119909 119910) = 1119899119899minus1
sum119894=1
1003816100381610038161003816119891119894 (119909 119910) minus 119891119894+1 (119909 119910)1003816100381610038161003816 (12)
where Δ(x y) is the average temporal variation 119891119894(xy) is apixel intensity at the location (x y) in the ith frame If Δ(x y)gtthr (experimentally determined threshold) then a movingpixel is fire
3 Results and Performance Analysis
Tomeasure the performance of the proposedmodel 6 videoswere collected from Internet 3 of them are available at(httpwwwultimatechasecom) Four of these videos wereactual fire and two were fire-color objects The algorithmwas implemented using MATLAB (R2017a) and tested on anIntel core i7 297 GHz PC 8GB-RAM PC Figure 6 showsthe variety of forest fire condition videos used in the test Atrue-positive was counted if an image frame had fire pixelsand it was determined by the proposed model to be fire Incontrast false-positive was counted if the image frame hasno fire and the result was determined as a fire Table 1 shows
Mathematical Problems in Engineering 5
(a) (b) (c) (d) (e) (f)
Figure 5 Applying the rules (7)-(11) to input images (a) original RGB images (b) binary images using rule (7) (c) binary images using rule(8) (d) binary images using rule (9) (e) binary images using rule (10) (f) binary images using rules (7) through (11)
Video_1 Video_2 Video_3 Video_4 Video_5 Video_6
Figure 6 Tested forest fire videos
the true-positive false-positive percentage of true-positiveand percentage of false-positive for tested videos
The results in Table 1 show that the proposed methodhas achieved average true-positive percentage (TTP) up to9663 in the tested forest fire videos and 923 false-positiverate These results indicate the good performance of theproposed method in forest fire detection
31 Performance Evaluation To evaluate the proposedmethod comparison between some of the above-mentionedmethods and the proposed one was carried out All of thesemethodswere tested in data sets consisting of 500 images (300images of forest fire and 200 nonfire images) collected fromInternet Algorithmsrsquo performances were calculated using theevaluation metric F-score
311 F-Score TheF-score [12] is used in this study to evaluatethe performance of the fire detection algorithms For anydetection method there are four potential results If an imagehas fire pixels and it was determined by the method as firethen it is true-positive If the same image is determined tobe not a fire pixel by the algorithm it is false-negative Ifan image has no fire and it was detected by the method asno fire it is a true-negative but if it was detected as fireby the method it counts as a false-positive Fire detectionalgorithms are evaluated using the following equations
where F is F-score TP TN FP and FN are true-positivetrue-negative false-positive and false-negative respectivelyA highermethod F-scoremeans a better overall performanceTable 2 and Figure 7 show the comparison results
(i) TP-rate obtained TP divided by the total number offire images
(ii) TN-rate obtained TN divided by the total number ofnonfire images
(iii) FN-rate obtained FN divided by the total number offire images
(iv) FP-rate obtained FP divided by the total number ofnonfire images
Figure 6 shows the F-score of the three methods Theproposedmethod F-score is higher than the existing methodsdescribed in [6 7] this indicates that the proposed methodperforms better than the existing methods
4 Conclusions
This study proposes an effective forest fire detection methodusing image processing techniques includingmovement con-taining region detection based on background subtractionand color segmentation The algorithm uses YCbCr colorspace which is better in separating the luminance from thechrominance and has good detection rate five fire detectionrules are applied to detect the fire The performance of theproposed algorithm is tested on data set consisting of 6 videos
collected from Internet four of which were actual fire videoswhile two were fire-like objects videos TP-rate and TN-ratewere calculatedThe results show that the proposed algorithmachieves good detection rates These results indicate that theproposed method is accurate and can be used in automaticforest fire-alarm systems
For future work the system could be improved by usinga combination of rules of different color spaces howeverthe challenge is selecting the right rules from different colorspaces to build the method
Data Availability
The data consists of 6 videos 3 of them are available at(httpwwwultimatechasecom) with license and the other3 videos are collected randomly from Internet
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
The work is supported by Fundamental Research Funds forthe Central Universities (2572017PZ10)
References
[1] B Ko and S Kwak ldquoSurvey of computer visionbased naturaldisaster warning systemsrdquo Optical Engineering vol 51 no 7Article ID 070901 2012
[2] J R Martinez-de Dios B C Arrue A Ollero L Merino andF Gomez-Rodrıguez ldquoComputer vision techniques for forestfire perceptionrdquo Image and Vision Computing vol 26 no 4 pp550ndash562 2008
[3] Y Meng Y Deng and P Shi ldquoMapping Forest Wildfire Risk ofthe Worldrdquo in World Atlas of Natural Disaster Risk P Shi andR Kasperson Eds pp 261ndash275 Springer Berlin HeidelbergBerlin Germany 2015
[4] P M Hanamaraddi ldquoA Literature Study on Image Processingfor Forest Fire Detectionrdquo IJITR vol 4 pp 2695ndash2700 2016
[5] P Podrzaj and H Hashimoto ldquoIntelligent space as a frameworkfor fire detection and evacuationrdquo Fire Technology vol 44 no1 pp 65ndash76 2008
[6] C Thou-Ho W Ping-Hsueh and C Yung-Chuen ldquoAn earlyfire-detection method based on image processingrdquo in Proceed-ings of the 2004 International Conference on Image ProcessingICIP rsquo04 vol 3 pp 1707ndash1710 2004
[7] Y Wang and J Ye ldquoResearch on the algorithm of preventionforest fire disaster in the Poyang Lake Ecological EconomicZonerdquo AdvancedMaterials Research pp 5257ndash5260 2012
[8] A D Alzughaibi H A Hakami and Z Chaczko ldquoReviewof human motion detection based on background subtractiontechniquesrdquo International Journal of ComputerApplications vol122 2015
[9] C E Premal and S S Vinsley ldquoImage processing based forestfire detection using YCbCr colour modelrdquo in Proceedings of the2014 International Conference on Circuits Power and ComputingTechnologies ICCPCT 2014 pp 1229ndash1237 2014
[10] VVipin ldquoImage processing based forest fire detectionrdquo Interna-tional Journal of Emerging Technology and Advanced Engineer-ing vol 2 pp 87ndash95 2012
[11] L-H Chen and W-C Huang ldquoFire detection using spatial-temporal analysisrdquo in Proceedings of the World Congress onEngineering pp 3ndash5 2013
[12] T Fawcett ldquoROC graphs Notes and practical considerations forresearchersrdquoMachine Learning vol 31 pp 1ndash38 2004
Hindawiwwwhindawicom Volume 2018
MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Mathematical Problems in Engineering
Applied MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Probability and StatisticsHindawiwwwhindawicom Volume 2018
Journal of
Hindawiwwwhindawicom Volume 2018
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawiwwwhindawicom Volume 2018
OptimizationJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Engineering Mathematics
International Journal of
Hindawiwwwhindawicom Volume 2018
Operations ResearchAdvances in
Journal of
Hindawiwwwhindawicom Volume 2018
Function SpacesAbstract and Applied AnalysisHindawiwwwhindawicom Volume 2018
International Journal of Mathematics and Mathematical Sciences
Numerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisAdvances inAdvances in Discrete Dynamics in
Nature and SocietyHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Dierential EquationsInternational Journal of
Volume 2018
Hindawiwwwhindawicom Volume 2018
Decision SciencesAdvances in
Hindawiwwwhindawicom Volume 2018
AnalysisInternational Journal of
Hindawiwwwhindawicom Volume 2018
Stochastic AnalysisInternational Journal of
Submit your manuscripts atwwwhindawicom
2 Mathematical Problems in Engineering
Get image sequence
Apply background-subtraction Update background model
Moving Obj available
RGB to YCbCr convert
Apply fire detection rules
fire detected Set fire alarm
NO
NO
YES
YES
temporal variation
Figure 1 The proposed fire detection method
from false-positive alarms in case of moving fire-like objectsDios et al [2] presented an optical model used to detect forestfires and measure the properties of the fire such as flameheight fire front fire base width and flame inclination angleThis system is very good nevertheless it is very expensivebecause it consists of infrared cameras and other technologiessuch as GPS and telemetry sensors Yinglian et al [7]proposed forest fire disaster prevention algorithm based onimage processing This algorithm depends on fire and smokecolor properties to identify fire Yinglianrsquos algorithm is goodbut the smoke spreads quickly and it hasmany different colorswhich depend on the burning material thus the false alarmrate rises
In this paper a forest fire detection algorithm is proposedThe algorithm uses YCbCr color space since it effectively sep-arates luminance from chrominance and is able to separatehigh temperature fire center pixels because the fire at the hightemperature center region is white The final results showthat the proposed system has good detection rates and fewerfalse alarms which are the main crucial problems of the mostexisting algorithms
This paper is organized as follows Section 2 describes theproposed fire detection method Section 3 presents the resultsand computational complexity of the proposed algorithm
and Section 4 summarizes the work that has been carried outin this study and potential future direction
2 Materials and Methods
This section presents the proposed forest fire detectionalgorithm It consists of the following main stages the firststep receives the input video from the input device the secondstep applies movement containing region detection based onbackground subtraction (MRDB) the third step converts theinput image sequence from RGB to YCbCr color space andthe fourth step applies the fire detection rules and temporalvariation A fire alarm is activated if all detection conditionsare satisfied The proposed algorithm stages will be describedin detail Figure 1 shows the proposed algorithm flowchart
21 Movement Containing Region Detection Based on Back-ground Subtraction (MRDB) Detecting moving regions isa key factor in most of the video based fire detectionsystems because fire boundaries continuously fluctuate Sobackground subtraction is used to select candidate regions offire A pixel located at (x y) is supposed to be moving if thefollowing condition is satisfied [8]
1003816100381610038161003816119868119899 (119909 119910) minus 119861119899 (119909 119910)1003816100381610038161003816 gt 119905ℎ119903 (1)
Mathematical Problems in Engineering 3
(a) (b)
Figure 2 Example of MRDB (a) Frame before MRDB (b) frame after applying MRDB
(a) (Y) (Cb) (Cr)
Figure 3 (a) An RGB image and its (Y) (Cb) and (Cr) channels
119868119899(x y) represents the intensity value of the pixel at location(x y) in the nth gray-level for the current frame and Bn (xy) represent the background intensity value at the same pixellocation and thr is a threshold value experimentally set to 3The background is continuously updated using (2)
where 119861119899+1(x y) and 119861119899(x y) represent the intensity of thepixel value at location (x y) for the current and previousbackgrounds Figure 2 shows an example of MRDB
22 Converting RGB Images to YCbCr Due to the fact thatdifferent kinds of moving objects can be included after apply-ing background subtraction such as trees animals birds andpeople therefore images from the background subtractionstage are converted to YCbCr [9] to select candidate fireregions using (3) Figure 3 shows original RGB image (a) andYCbCr component The mean values of the YCbCr channelare then calculated using (4) (5) and (6)
where 119884119898119890119886119899 119862119887119898119890119886119899 and 119862119903119898119890119886119899 are the mean values for theYCbCr channels Y (x y) Cb (x y) and Cr (x y) are YCbCrchannel values for pixel at specific location (x y) and NlowastMis the total number of pixels
23 Fire Color Pixel Detection Rules In any fire image pixelsthe red color value is larger than green and green is largerthan blue as illustrated in Figure 4 (a) is a fire image and(b) is the RGB channels histogram for the same image Thisfact is represented in RGB color space as RgtGgtB and can beconverted to YCbCr using the flowing equations
Also the Y component value is greater than the mean Ycomponent of the same image and the Cb component issmaller than the mean Cb of same image while the Cr
4 Mathematical Problems in Engineering
(a)
0 50 100 150 200 250 3000
200
400
600
800
1000
1200
1400
1600
1800
Blue channelGreen channelRed channel
(b)
Figure 4 (a) RGB fire image and (b) the Histogram of the same image channels
component is greater than the mean Cr mean componentThis fact can be represented by the flow
where F(x y) can be any pixel on the image Y119898 Cb119898 andCr119898 are the mean values for Y Cb and Cr respectively
The Cb component as shown is predominantly ldquoblackrdquoand the Cr component is ldquowhiterdquo This idea can be repre-sented by the following equations
119865 (119909 119910) = 1 if 1003816100381610038161003816119862119887 (119909 119910) minus 119862119903 (119909 119910)1003816100381610038161003816 ge 1205910 119874119905ℎ119890119903119908119894119904119890 (10)
119865 (119909 119910)
= 1 (119862119887 (119909 119910) le 120)⋂ (119862119903 (119909 119910) ge 150)0 119874119905ℎ119890119903119908119894119904119890
(11)
where 120591 is a constant specified in [10] using receiver oper-ating characteristic (ROC) by applying different values of 120591in the range [1 100] To measure the ldquotrue detection raterdquoand ldquofalse detectionrdquo data sets consisting of 500 images (300of them being images of a forest fire 200 nonfire images)collected from Internet were used Only (10) was used withdifferent values of 120591 in the range [1 100] to detect binaryimages of the candidate fire region 120591 was selected as 120591 = 70resulting in a true detection rate of more than 90 and falsedetection of less than 40
Figure 5 shows the application of detection rules (7)through (11)
24 Temporal Variation Using color models alone is notenough to identify fire correctly because there are severalobjects that share the samefire color such as red leaves desertand other red moving objects The main difference betweenactual fire and the fire-color objects is the nature of theirmotion Shape and size of the flame are totally changeablebecause of burning materials and airflow thus it produceshigher temporal variation In contrast rigid bodiesrsquo motionproduces lower temporal variation Therefore it is possibleto differentiate between the fire pixels and the fire color Todetect a fire movement the difference between successiveframes was analyzed Suppose a video sequence consisting ofn frames and the average temporal variation is defined as [11]
Δ (119909 119910) = 1119899119899minus1
sum119894=1
1003816100381610038161003816119891119894 (119909 119910) minus 119891119894+1 (119909 119910)1003816100381610038161003816 (12)
where Δ(x y) is the average temporal variation 119891119894(xy) is apixel intensity at the location (x y) in the ith frame If Δ(x y)gtthr (experimentally determined threshold) then a movingpixel is fire
3 Results and Performance Analysis
Tomeasure the performance of the proposedmodel 6 videoswere collected from Internet 3 of them are available at(httpwwwultimatechasecom) Four of these videos wereactual fire and two were fire-color objects The algorithmwas implemented using MATLAB (R2017a) and tested on anIntel core i7 297 GHz PC 8GB-RAM PC Figure 6 showsthe variety of forest fire condition videos used in the test Atrue-positive was counted if an image frame had fire pixelsand it was determined by the proposed model to be fire Incontrast false-positive was counted if the image frame hasno fire and the result was determined as a fire Table 1 shows
Mathematical Problems in Engineering 5
(a) (b) (c) (d) (e) (f)
Figure 5 Applying the rules (7)-(11) to input images (a) original RGB images (b) binary images using rule (7) (c) binary images using rule(8) (d) binary images using rule (9) (e) binary images using rule (10) (f) binary images using rules (7) through (11)
Video_1 Video_2 Video_3 Video_4 Video_5 Video_6
Figure 6 Tested forest fire videos
the true-positive false-positive percentage of true-positiveand percentage of false-positive for tested videos
The results in Table 1 show that the proposed methodhas achieved average true-positive percentage (TTP) up to9663 in the tested forest fire videos and 923 false-positiverate These results indicate the good performance of theproposed method in forest fire detection
31 Performance Evaluation To evaluate the proposedmethod comparison between some of the above-mentionedmethods and the proposed one was carried out All of thesemethodswere tested in data sets consisting of 500 images (300images of forest fire and 200 nonfire images) collected fromInternet Algorithmsrsquo performances were calculated using theevaluation metric F-score
311 F-Score TheF-score [12] is used in this study to evaluatethe performance of the fire detection algorithms For anydetection method there are four potential results If an imagehas fire pixels and it was determined by the method as firethen it is true-positive If the same image is determined tobe not a fire pixel by the algorithm it is false-negative Ifan image has no fire and it was detected by the method asno fire it is a true-negative but if it was detected as fireby the method it counts as a false-positive Fire detectionalgorithms are evaluated using the following equations
where F is F-score TP TN FP and FN are true-positivetrue-negative false-positive and false-negative respectivelyA highermethod F-scoremeans a better overall performanceTable 2 and Figure 7 show the comparison results
(i) TP-rate obtained TP divided by the total number offire images
(ii) TN-rate obtained TN divided by the total number ofnonfire images
(iii) FN-rate obtained FN divided by the total number offire images
(iv) FP-rate obtained FP divided by the total number ofnonfire images
Figure 6 shows the F-score of the three methods Theproposedmethod F-score is higher than the existing methodsdescribed in [6 7] this indicates that the proposed methodperforms better than the existing methods
4 Conclusions
This study proposes an effective forest fire detection methodusing image processing techniques includingmovement con-taining region detection based on background subtractionand color segmentation The algorithm uses YCbCr colorspace which is better in separating the luminance from thechrominance and has good detection rate five fire detectionrules are applied to detect the fire The performance of theproposed algorithm is tested on data set consisting of 6 videos
collected from Internet four of which were actual fire videoswhile two were fire-like objects videos TP-rate and TN-ratewere calculatedThe results show that the proposed algorithmachieves good detection rates These results indicate that theproposed method is accurate and can be used in automaticforest fire-alarm systems
For future work the system could be improved by usinga combination of rules of different color spaces howeverthe challenge is selecting the right rules from different colorspaces to build the method
Data Availability
The data consists of 6 videos 3 of them are available at(httpwwwultimatechasecom) with license and the other3 videos are collected randomly from Internet
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
The work is supported by Fundamental Research Funds forthe Central Universities (2572017PZ10)
References
[1] B Ko and S Kwak ldquoSurvey of computer visionbased naturaldisaster warning systemsrdquo Optical Engineering vol 51 no 7Article ID 070901 2012
[2] J R Martinez-de Dios B C Arrue A Ollero L Merino andF Gomez-Rodrıguez ldquoComputer vision techniques for forestfire perceptionrdquo Image and Vision Computing vol 26 no 4 pp550ndash562 2008
[3] Y Meng Y Deng and P Shi ldquoMapping Forest Wildfire Risk ofthe Worldrdquo in World Atlas of Natural Disaster Risk P Shi andR Kasperson Eds pp 261ndash275 Springer Berlin HeidelbergBerlin Germany 2015
[4] P M Hanamaraddi ldquoA Literature Study on Image Processingfor Forest Fire Detectionrdquo IJITR vol 4 pp 2695ndash2700 2016
[5] P Podrzaj and H Hashimoto ldquoIntelligent space as a frameworkfor fire detection and evacuationrdquo Fire Technology vol 44 no1 pp 65ndash76 2008
[6] C Thou-Ho W Ping-Hsueh and C Yung-Chuen ldquoAn earlyfire-detection method based on image processingrdquo in Proceed-ings of the 2004 International Conference on Image ProcessingICIP rsquo04 vol 3 pp 1707ndash1710 2004
[7] Y Wang and J Ye ldquoResearch on the algorithm of preventionforest fire disaster in the Poyang Lake Ecological EconomicZonerdquo AdvancedMaterials Research pp 5257ndash5260 2012
[8] A D Alzughaibi H A Hakami and Z Chaczko ldquoReviewof human motion detection based on background subtractiontechniquesrdquo International Journal of ComputerApplications vol122 2015
[9] C E Premal and S S Vinsley ldquoImage processing based forestfire detection using YCbCr colour modelrdquo in Proceedings of the2014 International Conference on Circuits Power and ComputingTechnologies ICCPCT 2014 pp 1229ndash1237 2014
[10] VVipin ldquoImage processing based forest fire detectionrdquo Interna-tional Journal of Emerging Technology and Advanced Engineer-ing vol 2 pp 87ndash95 2012
[11] L-H Chen and W-C Huang ldquoFire detection using spatial-temporal analysisrdquo in Proceedings of the World Congress onEngineering pp 3ndash5 2013
[12] T Fawcett ldquoROC graphs Notes and practical considerations forresearchersrdquoMachine Learning vol 31 pp 1ndash38 2004
Hindawiwwwhindawicom Volume 2018
MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Mathematical Problems in Engineering
Applied MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Probability and StatisticsHindawiwwwhindawicom Volume 2018
Journal of
Hindawiwwwhindawicom Volume 2018
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawiwwwhindawicom Volume 2018
OptimizationJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Engineering Mathematics
International Journal of
Hindawiwwwhindawicom Volume 2018
Operations ResearchAdvances in
Journal of
Hindawiwwwhindawicom Volume 2018
Function SpacesAbstract and Applied AnalysisHindawiwwwhindawicom Volume 2018
International Journal of Mathematics and Mathematical Sciences
Numerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisAdvances inAdvances in Discrete Dynamics in
Nature and SocietyHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Dierential EquationsInternational Journal of
Volume 2018
Hindawiwwwhindawicom Volume 2018
Decision SciencesAdvances in
Hindawiwwwhindawicom Volume 2018
AnalysisInternational Journal of
Hindawiwwwhindawicom Volume 2018
Stochastic AnalysisInternational Journal of
Submit your manuscripts atwwwhindawicom
Mathematical Problems in Engineering 3
(a) (b)
Figure 2 Example of MRDB (a) Frame before MRDB (b) frame after applying MRDB
(a) (Y) (Cb) (Cr)
Figure 3 (a) An RGB image and its (Y) (Cb) and (Cr) channels
119868119899(x y) represents the intensity value of the pixel at location(x y) in the nth gray-level for the current frame and Bn (xy) represent the background intensity value at the same pixellocation and thr is a threshold value experimentally set to 3The background is continuously updated using (2)
where 119861119899+1(x y) and 119861119899(x y) represent the intensity of thepixel value at location (x y) for the current and previousbackgrounds Figure 2 shows an example of MRDB
22 Converting RGB Images to YCbCr Due to the fact thatdifferent kinds of moving objects can be included after apply-ing background subtraction such as trees animals birds andpeople therefore images from the background subtractionstage are converted to YCbCr [9] to select candidate fireregions using (3) Figure 3 shows original RGB image (a) andYCbCr component The mean values of the YCbCr channelare then calculated using (4) (5) and (6)
where 119884119898119890119886119899 119862119887119898119890119886119899 and 119862119903119898119890119886119899 are the mean values for theYCbCr channels Y (x y) Cb (x y) and Cr (x y) are YCbCrchannel values for pixel at specific location (x y) and NlowastMis the total number of pixels
23 Fire Color Pixel Detection Rules In any fire image pixelsthe red color value is larger than green and green is largerthan blue as illustrated in Figure 4 (a) is a fire image and(b) is the RGB channels histogram for the same image Thisfact is represented in RGB color space as RgtGgtB and can beconverted to YCbCr using the flowing equations
Also the Y component value is greater than the mean Ycomponent of the same image and the Cb component issmaller than the mean Cb of same image while the Cr
4 Mathematical Problems in Engineering
(a)
0 50 100 150 200 250 3000
200
400
600
800
1000
1200
1400
1600
1800
Blue channelGreen channelRed channel
(b)
Figure 4 (a) RGB fire image and (b) the Histogram of the same image channels
component is greater than the mean Cr mean componentThis fact can be represented by the flow
where F(x y) can be any pixel on the image Y119898 Cb119898 andCr119898 are the mean values for Y Cb and Cr respectively
The Cb component as shown is predominantly ldquoblackrdquoand the Cr component is ldquowhiterdquo This idea can be repre-sented by the following equations
119865 (119909 119910) = 1 if 1003816100381610038161003816119862119887 (119909 119910) minus 119862119903 (119909 119910)1003816100381610038161003816 ge 1205910 119874119905ℎ119890119903119908119894119904119890 (10)
119865 (119909 119910)
= 1 (119862119887 (119909 119910) le 120)⋂ (119862119903 (119909 119910) ge 150)0 119874119905ℎ119890119903119908119894119904119890
(11)
where 120591 is a constant specified in [10] using receiver oper-ating characteristic (ROC) by applying different values of 120591in the range [1 100] To measure the ldquotrue detection raterdquoand ldquofalse detectionrdquo data sets consisting of 500 images (300of them being images of a forest fire 200 nonfire images)collected from Internet were used Only (10) was used withdifferent values of 120591 in the range [1 100] to detect binaryimages of the candidate fire region 120591 was selected as 120591 = 70resulting in a true detection rate of more than 90 and falsedetection of less than 40
Figure 5 shows the application of detection rules (7)through (11)
24 Temporal Variation Using color models alone is notenough to identify fire correctly because there are severalobjects that share the samefire color such as red leaves desertand other red moving objects The main difference betweenactual fire and the fire-color objects is the nature of theirmotion Shape and size of the flame are totally changeablebecause of burning materials and airflow thus it produceshigher temporal variation In contrast rigid bodiesrsquo motionproduces lower temporal variation Therefore it is possibleto differentiate between the fire pixels and the fire color Todetect a fire movement the difference between successiveframes was analyzed Suppose a video sequence consisting ofn frames and the average temporal variation is defined as [11]
Δ (119909 119910) = 1119899119899minus1
sum119894=1
1003816100381610038161003816119891119894 (119909 119910) minus 119891119894+1 (119909 119910)1003816100381610038161003816 (12)
where Δ(x y) is the average temporal variation 119891119894(xy) is apixel intensity at the location (x y) in the ith frame If Δ(x y)gtthr (experimentally determined threshold) then a movingpixel is fire
3 Results and Performance Analysis
Tomeasure the performance of the proposedmodel 6 videoswere collected from Internet 3 of them are available at(httpwwwultimatechasecom) Four of these videos wereactual fire and two were fire-color objects The algorithmwas implemented using MATLAB (R2017a) and tested on anIntel core i7 297 GHz PC 8GB-RAM PC Figure 6 showsthe variety of forest fire condition videos used in the test Atrue-positive was counted if an image frame had fire pixelsand it was determined by the proposed model to be fire Incontrast false-positive was counted if the image frame hasno fire and the result was determined as a fire Table 1 shows
Mathematical Problems in Engineering 5
(a) (b) (c) (d) (e) (f)
Figure 5 Applying the rules (7)-(11) to input images (a) original RGB images (b) binary images using rule (7) (c) binary images using rule(8) (d) binary images using rule (9) (e) binary images using rule (10) (f) binary images using rules (7) through (11)
Video_1 Video_2 Video_3 Video_4 Video_5 Video_6
Figure 6 Tested forest fire videos
the true-positive false-positive percentage of true-positiveand percentage of false-positive for tested videos
The results in Table 1 show that the proposed methodhas achieved average true-positive percentage (TTP) up to9663 in the tested forest fire videos and 923 false-positiverate These results indicate the good performance of theproposed method in forest fire detection
31 Performance Evaluation To evaluate the proposedmethod comparison between some of the above-mentionedmethods and the proposed one was carried out All of thesemethodswere tested in data sets consisting of 500 images (300images of forest fire and 200 nonfire images) collected fromInternet Algorithmsrsquo performances were calculated using theevaluation metric F-score
311 F-Score TheF-score [12] is used in this study to evaluatethe performance of the fire detection algorithms For anydetection method there are four potential results If an imagehas fire pixels and it was determined by the method as firethen it is true-positive If the same image is determined tobe not a fire pixel by the algorithm it is false-negative Ifan image has no fire and it was detected by the method asno fire it is a true-negative but if it was detected as fireby the method it counts as a false-positive Fire detectionalgorithms are evaluated using the following equations
where F is F-score TP TN FP and FN are true-positivetrue-negative false-positive and false-negative respectivelyA highermethod F-scoremeans a better overall performanceTable 2 and Figure 7 show the comparison results
(i) TP-rate obtained TP divided by the total number offire images
(ii) TN-rate obtained TN divided by the total number ofnonfire images
(iii) FN-rate obtained FN divided by the total number offire images
(iv) FP-rate obtained FP divided by the total number ofnonfire images
Figure 6 shows the F-score of the three methods Theproposedmethod F-score is higher than the existing methodsdescribed in [6 7] this indicates that the proposed methodperforms better than the existing methods
4 Conclusions
This study proposes an effective forest fire detection methodusing image processing techniques includingmovement con-taining region detection based on background subtractionand color segmentation The algorithm uses YCbCr colorspace which is better in separating the luminance from thechrominance and has good detection rate five fire detectionrules are applied to detect the fire The performance of theproposed algorithm is tested on data set consisting of 6 videos
collected from Internet four of which were actual fire videoswhile two were fire-like objects videos TP-rate and TN-ratewere calculatedThe results show that the proposed algorithmachieves good detection rates These results indicate that theproposed method is accurate and can be used in automaticforest fire-alarm systems
For future work the system could be improved by usinga combination of rules of different color spaces howeverthe challenge is selecting the right rules from different colorspaces to build the method
Data Availability
The data consists of 6 videos 3 of them are available at(httpwwwultimatechasecom) with license and the other3 videos are collected randomly from Internet
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
The work is supported by Fundamental Research Funds forthe Central Universities (2572017PZ10)
References
[1] B Ko and S Kwak ldquoSurvey of computer visionbased naturaldisaster warning systemsrdquo Optical Engineering vol 51 no 7Article ID 070901 2012
[2] J R Martinez-de Dios B C Arrue A Ollero L Merino andF Gomez-Rodrıguez ldquoComputer vision techniques for forestfire perceptionrdquo Image and Vision Computing vol 26 no 4 pp550ndash562 2008
[3] Y Meng Y Deng and P Shi ldquoMapping Forest Wildfire Risk ofthe Worldrdquo in World Atlas of Natural Disaster Risk P Shi andR Kasperson Eds pp 261ndash275 Springer Berlin HeidelbergBerlin Germany 2015
[4] P M Hanamaraddi ldquoA Literature Study on Image Processingfor Forest Fire Detectionrdquo IJITR vol 4 pp 2695ndash2700 2016
[5] P Podrzaj and H Hashimoto ldquoIntelligent space as a frameworkfor fire detection and evacuationrdquo Fire Technology vol 44 no1 pp 65ndash76 2008
[6] C Thou-Ho W Ping-Hsueh and C Yung-Chuen ldquoAn earlyfire-detection method based on image processingrdquo in Proceed-ings of the 2004 International Conference on Image ProcessingICIP rsquo04 vol 3 pp 1707ndash1710 2004
[7] Y Wang and J Ye ldquoResearch on the algorithm of preventionforest fire disaster in the Poyang Lake Ecological EconomicZonerdquo AdvancedMaterials Research pp 5257ndash5260 2012
[8] A D Alzughaibi H A Hakami and Z Chaczko ldquoReviewof human motion detection based on background subtractiontechniquesrdquo International Journal of ComputerApplications vol122 2015
[9] C E Premal and S S Vinsley ldquoImage processing based forestfire detection using YCbCr colour modelrdquo in Proceedings of the2014 International Conference on Circuits Power and ComputingTechnologies ICCPCT 2014 pp 1229ndash1237 2014
[10] VVipin ldquoImage processing based forest fire detectionrdquo Interna-tional Journal of Emerging Technology and Advanced Engineer-ing vol 2 pp 87ndash95 2012
[11] L-H Chen and W-C Huang ldquoFire detection using spatial-temporal analysisrdquo in Proceedings of the World Congress onEngineering pp 3ndash5 2013
[12] T Fawcett ldquoROC graphs Notes and practical considerations forresearchersrdquoMachine Learning vol 31 pp 1ndash38 2004
Hindawiwwwhindawicom Volume 2018
MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Mathematical Problems in Engineering
Applied MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Probability and StatisticsHindawiwwwhindawicom Volume 2018
Journal of
Hindawiwwwhindawicom Volume 2018
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawiwwwhindawicom Volume 2018
OptimizationJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Engineering Mathematics
International Journal of
Hindawiwwwhindawicom Volume 2018
Operations ResearchAdvances in
Journal of
Hindawiwwwhindawicom Volume 2018
Function SpacesAbstract and Applied AnalysisHindawiwwwhindawicom Volume 2018
International Journal of Mathematics and Mathematical Sciences
where F(x y) can be any pixel on the image Y119898 Cb119898 andCr119898 are the mean values for Y Cb and Cr respectively
The Cb component as shown is predominantly ldquoblackrdquoand the Cr component is ldquowhiterdquo This idea can be repre-sented by the following equations
119865 (119909 119910) = 1 if 1003816100381610038161003816119862119887 (119909 119910) minus 119862119903 (119909 119910)1003816100381610038161003816 ge 1205910 119874119905ℎ119890119903119908119894119904119890 (10)
119865 (119909 119910)
= 1 (119862119887 (119909 119910) le 120)⋂ (119862119903 (119909 119910) ge 150)0 119874119905ℎ119890119903119908119894119904119890
(11)
where 120591 is a constant specified in [10] using receiver oper-ating characteristic (ROC) by applying different values of 120591in the range [1 100] To measure the ldquotrue detection raterdquoand ldquofalse detectionrdquo data sets consisting of 500 images (300of them being images of a forest fire 200 nonfire images)collected from Internet were used Only (10) was used withdifferent values of 120591 in the range [1 100] to detect binaryimages of the candidate fire region 120591 was selected as 120591 = 70resulting in a true detection rate of more than 90 and falsedetection of less than 40
Figure 5 shows the application of detection rules (7)through (11)
24 Temporal Variation Using color models alone is notenough to identify fire correctly because there are severalobjects that share the samefire color such as red leaves desertand other red moving objects The main difference betweenactual fire and the fire-color objects is the nature of theirmotion Shape and size of the flame are totally changeablebecause of burning materials and airflow thus it produceshigher temporal variation In contrast rigid bodiesrsquo motionproduces lower temporal variation Therefore it is possibleto differentiate between the fire pixels and the fire color Todetect a fire movement the difference between successiveframes was analyzed Suppose a video sequence consisting ofn frames and the average temporal variation is defined as [11]
Δ (119909 119910) = 1119899119899minus1
sum119894=1
1003816100381610038161003816119891119894 (119909 119910) minus 119891119894+1 (119909 119910)1003816100381610038161003816 (12)
where Δ(x y) is the average temporal variation 119891119894(xy) is apixel intensity at the location (x y) in the ith frame If Δ(x y)gtthr (experimentally determined threshold) then a movingpixel is fire
3 Results and Performance Analysis
Tomeasure the performance of the proposedmodel 6 videoswere collected from Internet 3 of them are available at(httpwwwultimatechasecom) Four of these videos wereactual fire and two were fire-color objects The algorithmwas implemented using MATLAB (R2017a) and tested on anIntel core i7 297 GHz PC 8GB-RAM PC Figure 6 showsthe variety of forest fire condition videos used in the test Atrue-positive was counted if an image frame had fire pixelsand it was determined by the proposed model to be fire Incontrast false-positive was counted if the image frame hasno fire and the result was determined as a fire Table 1 shows
Mathematical Problems in Engineering 5
(a) (b) (c) (d) (e) (f)
Figure 5 Applying the rules (7)-(11) to input images (a) original RGB images (b) binary images using rule (7) (c) binary images using rule(8) (d) binary images using rule (9) (e) binary images using rule (10) (f) binary images using rules (7) through (11)
Video_1 Video_2 Video_3 Video_4 Video_5 Video_6
Figure 6 Tested forest fire videos
the true-positive false-positive percentage of true-positiveand percentage of false-positive for tested videos
The results in Table 1 show that the proposed methodhas achieved average true-positive percentage (TTP) up to9663 in the tested forest fire videos and 923 false-positiverate These results indicate the good performance of theproposed method in forest fire detection
31 Performance Evaluation To evaluate the proposedmethod comparison between some of the above-mentionedmethods and the proposed one was carried out All of thesemethodswere tested in data sets consisting of 500 images (300images of forest fire and 200 nonfire images) collected fromInternet Algorithmsrsquo performances were calculated using theevaluation metric F-score
311 F-Score TheF-score [12] is used in this study to evaluatethe performance of the fire detection algorithms For anydetection method there are four potential results If an imagehas fire pixels and it was determined by the method as firethen it is true-positive If the same image is determined tobe not a fire pixel by the algorithm it is false-negative Ifan image has no fire and it was detected by the method asno fire it is a true-negative but if it was detected as fireby the method it counts as a false-positive Fire detectionalgorithms are evaluated using the following equations
where F is F-score TP TN FP and FN are true-positivetrue-negative false-positive and false-negative respectivelyA highermethod F-scoremeans a better overall performanceTable 2 and Figure 7 show the comparison results
(i) TP-rate obtained TP divided by the total number offire images
(ii) TN-rate obtained TN divided by the total number ofnonfire images
(iii) FN-rate obtained FN divided by the total number offire images
(iv) FP-rate obtained FP divided by the total number ofnonfire images
Figure 6 shows the F-score of the three methods Theproposedmethod F-score is higher than the existing methodsdescribed in [6 7] this indicates that the proposed methodperforms better than the existing methods
4 Conclusions
This study proposes an effective forest fire detection methodusing image processing techniques includingmovement con-taining region detection based on background subtractionand color segmentation The algorithm uses YCbCr colorspace which is better in separating the luminance from thechrominance and has good detection rate five fire detectionrules are applied to detect the fire The performance of theproposed algorithm is tested on data set consisting of 6 videos
collected from Internet four of which were actual fire videoswhile two were fire-like objects videos TP-rate and TN-ratewere calculatedThe results show that the proposed algorithmachieves good detection rates These results indicate that theproposed method is accurate and can be used in automaticforest fire-alarm systems
For future work the system could be improved by usinga combination of rules of different color spaces howeverthe challenge is selecting the right rules from different colorspaces to build the method
Data Availability
The data consists of 6 videos 3 of them are available at(httpwwwultimatechasecom) with license and the other3 videos are collected randomly from Internet
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
The work is supported by Fundamental Research Funds forthe Central Universities (2572017PZ10)
References
[1] B Ko and S Kwak ldquoSurvey of computer visionbased naturaldisaster warning systemsrdquo Optical Engineering vol 51 no 7Article ID 070901 2012
[2] J R Martinez-de Dios B C Arrue A Ollero L Merino andF Gomez-Rodrıguez ldquoComputer vision techniques for forestfire perceptionrdquo Image and Vision Computing vol 26 no 4 pp550ndash562 2008
[3] Y Meng Y Deng and P Shi ldquoMapping Forest Wildfire Risk ofthe Worldrdquo in World Atlas of Natural Disaster Risk P Shi andR Kasperson Eds pp 261ndash275 Springer Berlin HeidelbergBerlin Germany 2015
[4] P M Hanamaraddi ldquoA Literature Study on Image Processingfor Forest Fire Detectionrdquo IJITR vol 4 pp 2695ndash2700 2016
[5] P Podrzaj and H Hashimoto ldquoIntelligent space as a frameworkfor fire detection and evacuationrdquo Fire Technology vol 44 no1 pp 65ndash76 2008
[6] C Thou-Ho W Ping-Hsueh and C Yung-Chuen ldquoAn earlyfire-detection method based on image processingrdquo in Proceed-ings of the 2004 International Conference on Image ProcessingICIP rsquo04 vol 3 pp 1707ndash1710 2004
[7] Y Wang and J Ye ldquoResearch on the algorithm of preventionforest fire disaster in the Poyang Lake Ecological EconomicZonerdquo AdvancedMaterials Research pp 5257ndash5260 2012
[8] A D Alzughaibi H A Hakami and Z Chaczko ldquoReviewof human motion detection based on background subtractiontechniquesrdquo International Journal of ComputerApplications vol122 2015
[9] C E Premal and S S Vinsley ldquoImage processing based forestfire detection using YCbCr colour modelrdquo in Proceedings of the2014 International Conference on Circuits Power and ComputingTechnologies ICCPCT 2014 pp 1229ndash1237 2014
[10] VVipin ldquoImage processing based forest fire detectionrdquo Interna-tional Journal of Emerging Technology and Advanced Engineer-ing vol 2 pp 87ndash95 2012
[11] L-H Chen and W-C Huang ldquoFire detection using spatial-temporal analysisrdquo in Proceedings of the World Congress onEngineering pp 3ndash5 2013
[12] T Fawcett ldquoROC graphs Notes and practical considerations forresearchersrdquoMachine Learning vol 31 pp 1ndash38 2004
Hindawiwwwhindawicom Volume 2018
MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Mathematical Problems in Engineering
Applied MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Probability and StatisticsHindawiwwwhindawicom Volume 2018
Journal of
Hindawiwwwhindawicom Volume 2018
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawiwwwhindawicom Volume 2018
OptimizationJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Engineering Mathematics
International Journal of
Hindawiwwwhindawicom Volume 2018
Operations ResearchAdvances in
Journal of
Hindawiwwwhindawicom Volume 2018
Function SpacesAbstract and Applied AnalysisHindawiwwwhindawicom Volume 2018
International Journal of Mathematics and Mathematical Sciences
Numerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisAdvances inAdvances in Discrete Dynamics in
Nature and SocietyHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Dierential EquationsInternational Journal of
Volume 2018
Hindawiwwwhindawicom Volume 2018
Decision SciencesAdvances in
Hindawiwwwhindawicom Volume 2018
AnalysisInternational Journal of
Hindawiwwwhindawicom Volume 2018
Stochastic AnalysisInternational Journal of
Submit your manuscripts atwwwhindawicom
Mathematical Problems in Engineering 5
(a) (b) (c) (d) (e) (f)
Figure 5 Applying the rules (7)-(11) to input images (a) original RGB images (b) binary images using rule (7) (c) binary images using rule(8) (d) binary images using rule (9) (e) binary images using rule (10) (f) binary images using rules (7) through (11)
Video_1 Video_2 Video_3 Video_4 Video_5 Video_6
Figure 6 Tested forest fire videos
the true-positive false-positive percentage of true-positiveand percentage of false-positive for tested videos
The results in Table 1 show that the proposed methodhas achieved average true-positive percentage (TTP) up to9663 in the tested forest fire videos and 923 false-positiverate These results indicate the good performance of theproposed method in forest fire detection
31 Performance Evaluation To evaluate the proposedmethod comparison between some of the above-mentionedmethods and the proposed one was carried out All of thesemethodswere tested in data sets consisting of 500 images (300images of forest fire and 200 nonfire images) collected fromInternet Algorithmsrsquo performances were calculated using theevaluation metric F-score
311 F-Score TheF-score [12] is used in this study to evaluatethe performance of the fire detection algorithms For anydetection method there are four potential results If an imagehas fire pixels and it was determined by the method as firethen it is true-positive If the same image is determined tobe not a fire pixel by the algorithm it is false-negative Ifan image has no fire and it was detected by the method asno fire it is a true-negative but if it was detected as fireby the method it counts as a false-positive Fire detectionalgorithms are evaluated using the following equations
where F is F-score TP TN FP and FN are true-positivetrue-negative false-positive and false-negative respectivelyA highermethod F-scoremeans a better overall performanceTable 2 and Figure 7 show the comparison results
(i) TP-rate obtained TP divided by the total number offire images
(ii) TN-rate obtained TN divided by the total number ofnonfire images
(iii) FN-rate obtained FN divided by the total number offire images
(iv) FP-rate obtained FP divided by the total number ofnonfire images
Figure 6 shows the F-score of the three methods Theproposedmethod F-score is higher than the existing methodsdescribed in [6 7] this indicates that the proposed methodperforms better than the existing methods
4 Conclusions
This study proposes an effective forest fire detection methodusing image processing techniques includingmovement con-taining region detection based on background subtractionand color segmentation The algorithm uses YCbCr colorspace which is better in separating the luminance from thechrominance and has good detection rate five fire detectionrules are applied to detect the fire The performance of theproposed algorithm is tested on data set consisting of 6 videos
collected from Internet four of which were actual fire videoswhile two were fire-like objects videos TP-rate and TN-ratewere calculatedThe results show that the proposed algorithmachieves good detection rates These results indicate that theproposed method is accurate and can be used in automaticforest fire-alarm systems
For future work the system could be improved by usinga combination of rules of different color spaces howeverthe challenge is selecting the right rules from different colorspaces to build the method
Data Availability
The data consists of 6 videos 3 of them are available at(httpwwwultimatechasecom) with license and the other3 videos are collected randomly from Internet
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
The work is supported by Fundamental Research Funds forthe Central Universities (2572017PZ10)
References
[1] B Ko and S Kwak ldquoSurvey of computer visionbased naturaldisaster warning systemsrdquo Optical Engineering vol 51 no 7Article ID 070901 2012
[2] J R Martinez-de Dios B C Arrue A Ollero L Merino andF Gomez-Rodrıguez ldquoComputer vision techniques for forestfire perceptionrdquo Image and Vision Computing vol 26 no 4 pp550ndash562 2008
[3] Y Meng Y Deng and P Shi ldquoMapping Forest Wildfire Risk ofthe Worldrdquo in World Atlas of Natural Disaster Risk P Shi andR Kasperson Eds pp 261ndash275 Springer Berlin HeidelbergBerlin Germany 2015
[4] P M Hanamaraddi ldquoA Literature Study on Image Processingfor Forest Fire Detectionrdquo IJITR vol 4 pp 2695ndash2700 2016
[5] P Podrzaj and H Hashimoto ldquoIntelligent space as a frameworkfor fire detection and evacuationrdquo Fire Technology vol 44 no1 pp 65ndash76 2008
[6] C Thou-Ho W Ping-Hsueh and C Yung-Chuen ldquoAn earlyfire-detection method based on image processingrdquo in Proceed-ings of the 2004 International Conference on Image ProcessingICIP rsquo04 vol 3 pp 1707ndash1710 2004
[7] Y Wang and J Ye ldquoResearch on the algorithm of preventionforest fire disaster in the Poyang Lake Ecological EconomicZonerdquo AdvancedMaterials Research pp 5257ndash5260 2012
[8] A D Alzughaibi H A Hakami and Z Chaczko ldquoReviewof human motion detection based on background subtractiontechniquesrdquo International Journal of ComputerApplications vol122 2015
[9] C E Premal and S S Vinsley ldquoImage processing based forestfire detection using YCbCr colour modelrdquo in Proceedings of the2014 International Conference on Circuits Power and ComputingTechnologies ICCPCT 2014 pp 1229ndash1237 2014
[10] VVipin ldquoImage processing based forest fire detectionrdquo Interna-tional Journal of Emerging Technology and Advanced Engineer-ing vol 2 pp 87ndash95 2012
[11] L-H Chen and W-C Huang ldquoFire detection using spatial-temporal analysisrdquo in Proceedings of the World Congress onEngineering pp 3ndash5 2013
[12] T Fawcett ldquoROC graphs Notes and practical considerations forresearchersrdquoMachine Learning vol 31 pp 1ndash38 2004
Hindawiwwwhindawicom Volume 2018
MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Mathematical Problems in Engineering
Applied MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Probability and StatisticsHindawiwwwhindawicom Volume 2018
Journal of
Hindawiwwwhindawicom Volume 2018
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawiwwwhindawicom Volume 2018
OptimizationJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Engineering Mathematics
International Journal of
Hindawiwwwhindawicom Volume 2018
Operations ResearchAdvances in
Journal of
Hindawiwwwhindawicom Volume 2018
Function SpacesAbstract and Applied AnalysisHindawiwwwhindawicom Volume 2018
International Journal of Mathematics and Mathematical Sciences
collected from Internet four of which were actual fire videoswhile two were fire-like objects videos TP-rate and TN-ratewere calculatedThe results show that the proposed algorithmachieves good detection rates These results indicate that theproposed method is accurate and can be used in automaticforest fire-alarm systems
For future work the system could be improved by usinga combination of rules of different color spaces howeverthe challenge is selecting the right rules from different colorspaces to build the method
Data Availability
The data consists of 6 videos 3 of them are available at(httpwwwultimatechasecom) with license and the other3 videos are collected randomly from Internet
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
The work is supported by Fundamental Research Funds forthe Central Universities (2572017PZ10)
References
[1] B Ko and S Kwak ldquoSurvey of computer visionbased naturaldisaster warning systemsrdquo Optical Engineering vol 51 no 7Article ID 070901 2012
[2] J R Martinez-de Dios B C Arrue A Ollero L Merino andF Gomez-Rodrıguez ldquoComputer vision techniques for forestfire perceptionrdquo Image and Vision Computing vol 26 no 4 pp550ndash562 2008
[3] Y Meng Y Deng and P Shi ldquoMapping Forest Wildfire Risk ofthe Worldrdquo in World Atlas of Natural Disaster Risk P Shi andR Kasperson Eds pp 261ndash275 Springer Berlin HeidelbergBerlin Germany 2015
[4] P M Hanamaraddi ldquoA Literature Study on Image Processingfor Forest Fire Detectionrdquo IJITR vol 4 pp 2695ndash2700 2016
[5] P Podrzaj and H Hashimoto ldquoIntelligent space as a frameworkfor fire detection and evacuationrdquo Fire Technology vol 44 no1 pp 65ndash76 2008
[6] C Thou-Ho W Ping-Hsueh and C Yung-Chuen ldquoAn earlyfire-detection method based on image processingrdquo in Proceed-ings of the 2004 International Conference on Image ProcessingICIP rsquo04 vol 3 pp 1707ndash1710 2004
[7] Y Wang and J Ye ldquoResearch on the algorithm of preventionforest fire disaster in the Poyang Lake Ecological EconomicZonerdquo AdvancedMaterials Research pp 5257ndash5260 2012
[8] A D Alzughaibi H A Hakami and Z Chaczko ldquoReviewof human motion detection based on background subtractiontechniquesrdquo International Journal of ComputerApplications vol122 2015
[9] C E Premal and S S Vinsley ldquoImage processing based forestfire detection using YCbCr colour modelrdquo in Proceedings of the2014 International Conference on Circuits Power and ComputingTechnologies ICCPCT 2014 pp 1229ndash1237 2014
[10] VVipin ldquoImage processing based forest fire detectionrdquo Interna-tional Journal of Emerging Technology and Advanced Engineer-ing vol 2 pp 87ndash95 2012
[11] L-H Chen and W-C Huang ldquoFire detection using spatial-temporal analysisrdquo in Proceedings of the World Congress onEngineering pp 3ndash5 2013
[12] T Fawcett ldquoROC graphs Notes and practical considerations forresearchersrdquoMachine Learning vol 31 pp 1ndash38 2004
Hindawiwwwhindawicom Volume 2018
MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Mathematical Problems in Engineering
Applied MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Probability and StatisticsHindawiwwwhindawicom Volume 2018
Journal of
Hindawiwwwhindawicom Volume 2018
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawiwwwhindawicom Volume 2018
OptimizationJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Engineering Mathematics
International Journal of
Hindawiwwwhindawicom Volume 2018
Operations ResearchAdvances in
Journal of
Hindawiwwwhindawicom Volume 2018
Function SpacesAbstract and Applied AnalysisHindawiwwwhindawicom Volume 2018
International Journal of Mathematics and Mathematical Sciences
collected from Internet four of which were actual fire videoswhile two were fire-like objects videos TP-rate and TN-ratewere calculatedThe results show that the proposed algorithmachieves good detection rates These results indicate that theproposed method is accurate and can be used in automaticforest fire-alarm systems
For future work the system could be improved by usinga combination of rules of different color spaces howeverthe challenge is selecting the right rules from different colorspaces to build the method
Data Availability
The data consists of 6 videos 3 of them are available at(httpwwwultimatechasecom) with license and the other3 videos are collected randomly from Internet
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
The work is supported by Fundamental Research Funds forthe Central Universities (2572017PZ10)
References
[1] B Ko and S Kwak ldquoSurvey of computer visionbased naturaldisaster warning systemsrdquo Optical Engineering vol 51 no 7Article ID 070901 2012
[2] J R Martinez-de Dios B C Arrue A Ollero L Merino andF Gomez-Rodrıguez ldquoComputer vision techniques for forestfire perceptionrdquo Image and Vision Computing vol 26 no 4 pp550ndash562 2008
[3] Y Meng Y Deng and P Shi ldquoMapping Forest Wildfire Risk ofthe Worldrdquo in World Atlas of Natural Disaster Risk P Shi andR Kasperson Eds pp 261ndash275 Springer Berlin HeidelbergBerlin Germany 2015
[4] P M Hanamaraddi ldquoA Literature Study on Image Processingfor Forest Fire Detectionrdquo IJITR vol 4 pp 2695ndash2700 2016
[5] P Podrzaj and H Hashimoto ldquoIntelligent space as a frameworkfor fire detection and evacuationrdquo Fire Technology vol 44 no1 pp 65ndash76 2008
[6] C Thou-Ho W Ping-Hsueh and C Yung-Chuen ldquoAn earlyfire-detection method based on image processingrdquo in Proceed-ings of the 2004 International Conference on Image ProcessingICIP rsquo04 vol 3 pp 1707ndash1710 2004
[7] Y Wang and J Ye ldquoResearch on the algorithm of preventionforest fire disaster in the Poyang Lake Ecological EconomicZonerdquo AdvancedMaterials Research pp 5257ndash5260 2012
[8] A D Alzughaibi H A Hakami and Z Chaczko ldquoReviewof human motion detection based on background subtractiontechniquesrdquo International Journal of ComputerApplications vol122 2015
[9] C E Premal and S S Vinsley ldquoImage processing based forestfire detection using YCbCr colour modelrdquo in Proceedings of the2014 International Conference on Circuits Power and ComputingTechnologies ICCPCT 2014 pp 1229ndash1237 2014
[10] VVipin ldquoImage processing based forest fire detectionrdquo Interna-tional Journal of Emerging Technology and Advanced Engineer-ing vol 2 pp 87ndash95 2012
[11] L-H Chen and W-C Huang ldquoFire detection using spatial-temporal analysisrdquo in Proceedings of the World Congress onEngineering pp 3ndash5 2013
[12] T Fawcett ldquoROC graphs Notes and practical considerations forresearchersrdquoMachine Learning vol 31 pp 1ndash38 2004
Hindawiwwwhindawicom Volume 2018
MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Mathematical Problems in Engineering
Applied MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Probability and StatisticsHindawiwwwhindawicom Volume 2018
Journal of
Hindawiwwwhindawicom Volume 2018
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawiwwwhindawicom Volume 2018
OptimizationJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Engineering Mathematics
International Journal of
Hindawiwwwhindawicom Volume 2018
Operations ResearchAdvances in
Journal of
Hindawiwwwhindawicom Volume 2018
Function SpacesAbstract and Applied AnalysisHindawiwwwhindawicom Volume 2018
International Journal of Mathematics and Mathematical Sciences
collected from Internet four of which were actual fire videoswhile two were fire-like objects videos TP-rate and TN-ratewere calculatedThe results show that the proposed algorithmachieves good detection rates These results indicate that theproposed method is accurate and can be used in automaticforest fire-alarm systems
For future work the system could be improved by usinga combination of rules of different color spaces howeverthe challenge is selecting the right rules from different colorspaces to build the method
Data Availability
The data consists of 6 videos 3 of them are available at(httpwwwultimatechasecom) with license and the other3 videos are collected randomly from Internet
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
The work is supported by Fundamental Research Funds forthe Central Universities (2572017PZ10)
References
[1] B Ko and S Kwak ldquoSurvey of computer visionbased naturaldisaster warning systemsrdquo Optical Engineering vol 51 no 7Article ID 070901 2012
[2] J R Martinez-de Dios B C Arrue A Ollero L Merino andF Gomez-Rodrıguez ldquoComputer vision techniques for forestfire perceptionrdquo Image and Vision Computing vol 26 no 4 pp550ndash562 2008
[3] Y Meng Y Deng and P Shi ldquoMapping Forest Wildfire Risk ofthe Worldrdquo in World Atlas of Natural Disaster Risk P Shi andR Kasperson Eds pp 261ndash275 Springer Berlin HeidelbergBerlin Germany 2015
[4] P M Hanamaraddi ldquoA Literature Study on Image Processingfor Forest Fire Detectionrdquo IJITR vol 4 pp 2695ndash2700 2016
[5] P Podrzaj and H Hashimoto ldquoIntelligent space as a frameworkfor fire detection and evacuationrdquo Fire Technology vol 44 no1 pp 65ndash76 2008
[6] C Thou-Ho W Ping-Hsueh and C Yung-Chuen ldquoAn earlyfire-detection method based on image processingrdquo in Proceed-ings of the 2004 International Conference on Image ProcessingICIP rsquo04 vol 3 pp 1707ndash1710 2004
[7] Y Wang and J Ye ldquoResearch on the algorithm of preventionforest fire disaster in the Poyang Lake Ecological EconomicZonerdquo AdvancedMaterials Research pp 5257ndash5260 2012
[8] A D Alzughaibi H A Hakami and Z Chaczko ldquoReviewof human motion detection based on background subtractiontechniquesrdquo International Journal of ComputerApplications vol122 2015
[9] C E Premal and S S Vinsley ldquoImage processing based forestfire detection using YCbCr colour modelrdquo in Proceedings of the2014 International Conference on Circuits Power and ComputingTechnologies ICCPCT 2014 pp 1229ndash1237 2014
[10] VVipin ldquoImage processing based forest fire detectionrdquo Interna-tional Journal of Emerging Technology and Advanced Engineer-ing vol 2 pp 87ndash95 2012
[11] L-H Chen and W-C Huang ldquoFire detection using spatial-temporal analysisrdquo in Proceedings of the World Congress onEngineering pp 3ndash5 2013
[12] T Fawcett ldquoROC graphs Notes and practical considerations forresearchersrdquoMachine Learning vol 31 pp 1ndash38 2004
Hindawiwwwhindawicom Volume 2018
MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Mathematical Problems in Engineering
Applied MathematicsJournal of
Hindawiwwwhindawicom Volume 2018
Probability and StatisticsHindawiwwwhindawicom Volume 2018
Journal of
Hindawiwwwhindawicom Volume 2018
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawiwwwhindawicom Volume 2018
OptimizationJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Engineering Mathematics
International Journal of
Hindawiwwwhindawicom Volume 2018
Operations ResearchAdvances in
Journal of
Hindawiwwwhindawicom Volume 2018
Function SpacesAbstract and Applied AnalysisHindawiwwwhindawicom Volume 2018
International Journal of Mathematics and Mathematical Sciences
