Fire Detection method in video using Covariance descriptors

fire detection in video sequences using a generic colour model and fire detection in video sequences using statistical color model video based fire detection system
Dr.JohnGates Profile Pic
Published Date:09-11-2017
Your Website URL(Optional)
I.J. Intelligent Systems and Applications, 2017, 2, 42-48 Published Online February 2017 in MECS ( DOI: 10.5815/ijisa.2017.02.06 Video-based Flame Detection using LBP-based Descriptor: Influences of Classifiers Variety on Detection Efficiency Oleksii Maksymiv Lviv State University of Life Safety, Lviv, Ukraine E-mail: Taras Rak Lviv State University of Life Safety, Lviv, Ukraine E-mail: Dmytro Peleshko Lviv Polytechnic National University, Lviv, Ukraine E-mail: Abstract—Techniques to detect the flame at an early which is analyzed, average parameters of detection stage are necessary in order to prevent the fire and accuracy and recall were calculated. minimize the damage. The flame detection technique based on the physical sensor has limited disadvantages in detecting the fire early. This paper presents the results of II. RESEARCH BACKGROUND using local binary patterns for solving flames detecting To solve the problem of fire detecting was used a problem and proposes modifications to improve the variety of approaches. Conditionally, they can be divided quality of detector work. Experimentally found that using into the following groups: color and texture segmentation, support vector machines classifier with a kernel based on motion detection in images, additional flame attributes Gaussian radial basis functions shows the best results (flickering, sharp corners, etc.). compared to other SVM cores or classifier k-nearest There are a lot of research works devoted to color neighbors. segmentation, which can be divided by type of model used: RGB 1, HSV 2, YCbCr 3, CIE Lab 4. In Index Terms—Computer vision, fire detection, image this paper, we used color model YCbCr, which shows the analysis, local binary patterns, image segmentation. best quality detection rate - 99% 3. However, despite the good results, presented by researchers, this method characterizes by the high rate of false positives alarms, I. INTRODUCTION that arise from changes in lighting and required Fire detection from the video stream is considered to parameters setting for each facility separately. Obtained be difficult and not fully resolved the problem in results, in our view, were largely represented by small computer vision areas research. The majority of fire representative dataset on which the test was performed, detection methods in images are based on such features as which, in particular, taken into account during our color, texture or some another temporal characteristic. experimental studies. However, the flame as object of classification has a To lower the false alarm rate, in addition to different difficult structure with unstable parameters of movement, color models, some researchers also incorporated motion wide range of colors (depending on temperature and and fire flickering into their work. Rinsurongkawong et al. burning substances), variable contours and other dynamic 4 have proposed to use Lucas Kanade optical flow effects. All these factors greatly increase the number of algorithm for fire moving detect. Toreyin et al. 5 used incorrect classification. hybrid background estimation method to detect moving This article describes the process of fire detecting with objects. For checks flicker in flames he using 1-D usage of local binary patterns (LBP) texture descriptor temporal wavelet transform and color variation in fire and its modifications, aimed at improving the flame colored moving regions using 2 –D spatial wavelet detector efficiency in video stream dissemination. Further, transform. Although powerful, these techniques are for received features, support vector machine classifier computationally intensive. (SVM) applied with different variations of kernels and k- Problems of using various descriptors to identify fire in nearest neighbors (k-NN) algorithm. As a result of using images/video stream are lighted in works 6 (SIFT this approach, the time required to process the image descriptor), 7 (HOG descriptor), 8 (SURF descriptor). Copyright © 2017 MECS I.J. Intelligent Systems and Applications, 2017, 2, 42-48 Video-based Flame Detection using LBP-based Descriptor: Influences of 43 Classifiers Variety on Detection Efficiency Also, there are some studies of using LBP descriptors for interpolation was used. Fig. 2 shows pixels set for the task of flames identify 9, 10. However, these works different neighborhood P and R. are not sufficiently disclosed potential choice classifier, which, according to experimental results can positively effect on developed detection system. In particular, we proposed a number of modifications LBP algorithm, which allow improving the quality of flame detection. Fig.2. Neighborhood examples used to define a texture and calculate a local binary pattern with different value P and R III. THEORETICAL BASIS OF LBP DESCRIPTOR As noted above, there are various feature extraction It has also been experimentally established, that for techniques: HOG, Haar Wavelets, LBP etc. For improving descriptor performance and efficiency, developed system, we have decided to use LBP descriptor. following set of pixels along a circle: P = 8, R = 1.0 is the The reason for this choice was caused by number of most optimal choice for our tasks. advantages for fire detection: speed calculation, In work 12 was proposed to analyze the images using invariability to brightness change which preserve order not all binary patterns, but only those who may contain at ) ) ) , resistance to noise and most two bitwise transitions from "1" to "0" or vice versa. textures variations. The number of bitwise transitions, when the binary string Classical LBP operator, which was proposed in work is circular, gives a uniformity measure U of the pattern as 11, forms labels for the image pixels by thresholding follows: the 3 x 3 neighborhood of each pixel with the center value and considering the result as a binary number. Using a circular neighborhood and bilinearly interpolating values at non-integer pixel coordinates ∑ ( ) ) allow any radius and number of pixels in the neighborhood. Pixels that have a value greater than the central pixel (or equal), take the value "1", those that are (3) less central, taking the value "0". The results of the comparisons are stored in an 8-bit binary code, which ) ( ) ) describes the neighborhood pixels. Neighborhood examples used to define a texture and calculate a local ∑ ( ) ) binary pattern shown in Fig.1. (4) These templates are called uniform pattern and characterized by two advantages. First, we save a memory because the number of uniform LBP patterns is ) given by , leading to a much shorter histogram representation. Secondly, they define only important local textures, such as the end of lines, edges, corners, spots. Further LBP operator with uniform patterns we will defined as In the last step, after the LBP labeled image has been Fig.1. Example of computing LBP obtained, the LBP histogram ( ) is computed, which can Mathematically, this process can be defined as follows: be defined as: ∑ ) (1) ∑ ( ) ) where is a center pixel value and ( ) (5) ) – value of pixels P, which are along a circle of radius R (R 0), s(x) - sign function, which can be defined as follows: In many cases, the LBP histogram may have excessive information, and much of it is often irrelevant to the problem to be solved. Which is why it was decided to use ) (2) the method proposed in 13. So, the histogram bins are sorted by descending order. After that, there is only part A number of pixels (P) along a circle with radius R can of the histogram that includes a percentage and patterns be chosen arbitrarily. To calculate values in these points which are then submitted for classification. for different radius R and points number P, bilinear Copyright © 2017 MECS I.J. Intelligent Systems and Applications, 2017, 2, 42-48 44 Video-based Flame Detection using LBP-based Descriptor: Influences of Classifiers Variety on Detection Efficiency ) ) ) IV. PROPOSED METHOD (10) Developed flame/non-flame detector must meet the The mean value for Y, Cb and Cr are calculated as requirements of implicitly high-level object detection and good performance even on slow machines. In this paper, equation: to reduce the time required for verification area, which may cause a fire, we proposed the additional use of color ) ∑∑ ) segmentation and some spatial-temporal object characteristics. (11) The process of detecting area with fire in images based on the five basic principles described below: 1. Color segmentation; ) ∑∑ ) 2. Moving objects detecting; 3. Applying LBP to obtained area from the previous (12) steps; 4. Based on the received histograms from the previous method - doing classification; ) ∑∑ ) 5. Fit into bounding box regions which qualifier recognize as "flame" and go to next frame. (13) A. Color segmentation B. Moving objects detection The performance of the fire detection system critically depends on the effectiveness of fire pixel classifier. A An estimate of the background (often called a common approach to color segmentation area with fire is background model) is computed and evolved frame by using color model YCbCr. For this, on the test dataset frame: moving objects in the scene are detected by the doing interactive image segmentation to identify the difference between the current frame and the current boundaries of which can be areas with flames. Tyrgay background model. Initially formed current background Celik et al. 3, 14 most fully analyzed the depending model ) over time t, by splitting the input image quality of fire segmentation from the selected color ) . To update the background model creates a model. As a result, it was found that YCbCr has higher circular buffer with size n = k + 1 elements, in which quality in comparison with the RGB model. Following recorded pixel brightness in k frames, and the brightness equation describes the color conversion from RGB to of the pixel ) , which contained in the initial YCbCr: background model ). Value of the cyclical buffer arranged in ascending order, and searches the meaning, which is located in the middle of the cyclical buffer. Updating background model ) was doing by replacing its corresponding pixel on the value in middle element of cyclical buffer. (6) Getting frame difference is due to calculation difference ) ) ) current video frame from the updated binary mask and calculating two where Y - luminance component; Cb - the difference thresholds cyclical buffer expressions: between blue component and brightness Y; Cr - the difference between red component and brightness Y. Segmentation area with flames was based on rules: ) ( ) (14) ) ) ) ) ) ) (7) where - number of row and column number which is at the intersection pixel frame, - the number of elements cyclical buffer, P - element circular buffer that i ) ) is on position . ) ) ) (8) In the resulting binary mask ), a pixel difference ) is classified as belonging a moving object, if he presents on mask Th and located into three pixels’ mask 1 ) Th , otherwise the pixel is considered as background. 2 ) For pixels segmentation that can be attributed to ) (9) ) micromotions generated by background, moving objects ) in the resulting binary mask allocated by grouping them all connected pixels and calculates their Copyright © 2017 MECS I.J. Intelligent Systems and Applications, 2017, 2, 42-48 Video-based Flame Detection using LBP-based Descriptor: Influences of 45 Classifiers Variety on Detection Efficiency modules gradient by the expression: Based on the above, binary codes of upper and lower images in fig.3 represent the same texture. This, in turn, will reduce feature vectors dimension at half. ) ) √ ‖ ‖ ‖ ‖ (15) E. VLBP Despite the fact that discriminatory ability of NRLBP ) where ‖ ‖ ) ) , characterized by improved capacity and much lower bins ) storage requirements, in comparison with classical LBP, ) ‖ ‖ ) - gradients in it includes only local visual information. To build a horizontal and vertical directions, ‖ ‖ - vector norm. quality vision-based fire monitoring system, primarily important information about its temporal characteristics. C. LBP descriptor using For this, we decide to use Volume Local Binary Pattern Taking into account the combustion process, we have (VLBP), which is calculated on the difference between chosen some modifications of LBP descriptor which will frames. improve the efficiency of our flame detector on the one In this method, arbitrary pixel frame of video is hand and increase computation speed on the other. described as ) and its coordinates D. NRLBP determined in accordance ) and times moment . Target (central) pixels has described by equation (17): Classic LBP descriptor characterized by two disadvantages that are described in 16, 23. First of all, it ) (17) is sensitive to changes between background and foreground. As can be observed in fig. 3, the binary code where – time between sequence. of LBP descriptor in two images with flames are different Neighboring pixels are selected according to equation from each other. However, they describe the same spatial (18): characteristics of fire. Another disadvantage of classic LBP descriptor it is high demands on the bins calculation ( ) and storage. Thus, the original implementation ) (18) required 256 histogram bins, in – only 59 bins, which is also quite costly. To develop flame detection where P - the number of neighboring pixels. system which can work in real time it is necessary to Dynamic LBP operator, which will depend on the time minimize requirements for their calculation and storage. interval between successive frames , number of Taking into consideration this disadvantages we neighboring pixels P and the distance between the target propose to use Non-Redundant Local Binary Pattern and adjacent pixels R. (NRLBP) 17 to obtain visual information about flame presence. From Fig. 3 can be seen that LBP codes of two images complement each other (the sum will amount ∑ ). Mathematically NRLBP can be defined as follows: (19) ) ) where – determines function (2), arguments of which ) (16) is the difference ), whose number is (( ) 18. As a result, after operators using we obtained a histogram, which, in turn, is fed to the input classifier for further processing. F. Classification Selection of qualitative features descriptor undoubtedly plays a key role in developing vision-based fire detection system. However, after receiving features that describe the desired image, the equally important step is the rejection of so-called "extra" features around the set. Therefore, classifier selection is important task during image classification and requires more detailed study. G. Support Vector Machine Classification using support vector method is widely Fig.3. LBP codes for the flame with same structure, but different used in image recognition thanks to its empirically good inversion in background and foreground performance and moderate calculation complexity. General principles of SVM described in 19, 20. SVM Copyright © 2017 MECS I.J. Intelligent Systems and Applications, 2017, 2, 42-48 46 Video-based Flame Detection using LBP-based Descriptor: Influences of Classifiers Variety on Detection Efficiency becomes famous when using pixel maps as input; it gives objects which may look like a flame (bright clothing, accuracy comparable to sophisticated neural networks lights, garlands, etc.). with elaborated features in a handwriting recognition task. To train the classifier we used image dataset that was It is also being used for many applications, such as hand formed during the previous study 22. It includes 1876 writing analysis, face analysis and so forth, especially for images with fire and the 4634 images without fire. pattern classification and regression based applications B. Evaluation method 19. It should also be noted that the classification function Results of fire classification evaluated based on two (F) becomes: well-known techniques - precision and recall. High precision relates to a low false positive rate, and( 20 hi)g h recall relates to a low false negative rate. High scores for ) ( )) ) (20) both show that the classifier is returning accurate results (high precision), as well as returning a majority of all and expression ́) ) ́)) called core positive results (high recall). classifier. Classification accuracy can be improved by proper kernels selection. The most common kernels are 24: (21) )  polynomial (homogeneous): ́ ́) ; (22) )  polynomial (inhomogeneous): ́ ́ ) ; where TP - the number of true positives (correct system  radial basis function (RBF): ́) inference about area where were flames); FP - false- ‖ ́‖) ; positive operation (system result agrees with the presence  Gaussian radial basis function (RBF- of fire whereas ground truth agrees with the absence of ‖ ́‖ G): ́) ) fire); FN - false-negative operation (system result agree of absence of fire whereas ground truth agree of the  sigmoid function: ́) ́ ) presence of fire). almost for all k0 і c0 C. Experimental results To analyze the performance of the flame classifier we The results figures of above methods are presented in analyzed this kernels in work, results of which are Table I. As we can see, using SVM classifier with a described in Section V. kernel based on radial basis function showed the best H. k-nearest neighbors algorithm precision result. SVM classifier with Gaussian radial basis function showed the best recall result in fire k-nearest neighbors algorithm is a non-parametric detection task. SVM with polynomial kernel showed the method based on a similarity measure (e.g., distance best time information processing. k-nearest neighbors functions). If K = 1, then the case is simply assigned to algorithm shows the worse performance of time and the class of its nearest neighbor. classification efficiency. A main advantage of the k-NN algorithm is that it performs well with multi-modal classes because the basis Table 1. Results of using different classification methods of its decision is based on a small neighborhood of similar objects. Therefore, even if the target class is Method Precision Recall Time multi-modal, the algorithm can still lead to good accuracy. SVM- polynomial 81,3 % 93,5 % 0,009 However, a major disadvantage of the k-NN algorithm is kernel that it uses all the features equally in computing for SVM-RBF kernel 81,9 % 94,2 % 0,02 similarities 21, 25. SVM-RBFG kernel 81,6 % 94,3 % 0,03 It should be noted about some disadvantages of k-NN: SVM- sigmoid kernel 81,3 % 93,7 % 0,03 model can not be interpreted, it is computationally k-NN 80,7 % 92, 6 % 0.01 expensive to find the k nearest neighbors when the dataset is very large, performance depends on the number of dimensions that we have. VI. CONCLUSIONS For developed video-based flame detector, the most V. RESULTS important parameter is the recall of classification. Using SVM classifier with Gaussian RBF kernel shows the best A. Dataset recall and relatively high performance of accuracy and For test classifications quality we formed the own performance time. So, it’s the most better choice for dataset of video, which is divided into two categories: the solving the problem of flame detection in video. flame and without flame. A number of copies used for the More future research should be done to verify and category "flame" - 22, for categories in which there are improve the results of this paper. In future work, we will no flames - 34. In particular, the last category includes endeavor to incorporate fire reflection detection Copyright © 2017 MECS I.J. Intelligent Systems and Applications, 2017, 2, 42-48 Video-based Flame Detection using LBP-based Descriptor: Influences of 47 Classifiers Variety on Detection Efficiency algorithms into our current algorithm. REFERENCES Some results of false positive and true 1 T. Chen, P. Wu, and Y. Chiou, “An early fire-detection positive/negative detection shown in Fig. 4, 5, 6. method based on image processing,” ICIP ’04, pp. 1707– 1710, 2004. 2 X. Qi, J. Ebert “A Computer Vision-Based Method for Fire Detection in Color Videos,” International journal of imaging, no. 9, pp. 22–34, 2009. 3 Celik T., H. Demirel “Fire detection in video sequences using a generic color model,” Fire Safety Journal, no.2, pp. 147–158, 2009. 4 S. Rinsurongkawong, M. Ekpanyapong, and M. N. Dailey, “Fire detection for early fire alarm based on optical flow video processing,” in 9th Int. Conf. on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology Phetchaburi, pp. 1-4, 2012. 5 B.U. Toreyin, Y. Dedeogln, U. Gudukbay, and A.E. Cetin , “Computer vision based method for real time fire and flame detection,” Pattern Recognition Letters, vol. 27, no. 4, pp. 49-58, 2006. 6 L.Ying, W. Hui-qin, B. Zhen-Bo and X. Fei “Research on Large Space Building Fire Positioning Technology Using Video Surveillance Images,”pp. 1218–1222, 2013. 7 P. Gomes, P. Santana, J. Barata “Vision-based Approach Fig.4. Examples of correct detection (True positive) to Fire Detection,” International Journal of Advanced Robotic Systems, no. 11, pp. 1–12, 2014. 8 Jiang, B., Lu, Y., Li, X. and Lin, L. “Towards a solid solution of realtime fire and flame detection,” Multimedia Tools and Applications., 2014. 9 F. Sthevanie, H. Nugroho, F. Arif Yulianto “Visual-Based Fire Detection Using Local Binary Pattern-Three Orthogonal Planes,” CYBERNETICSCOM, 2013, pp. 155–159. 10 Y. Zhao, T. Guizhong “Fire video recognition based on flame and smoke characteristics”, Systems and Informatics (ICSAI), 2014, pp. 113 – 118. 11 T. Ojala, M. Pietikainen, and T. Maenpaa “Multiresolution gray-scale and rotation invariant texture classification with local binary patterns”, IEEE Transactions on Pattern Analysis and Machine Intelligence, 24(7):971–987, 2002. 12 S. Nigam, A. Khare “Integration of moment invariants and uniform local binary patterns for human activity recognition in video sequences,” Multimed Tools Appl, 2015, pp. 1–30. 13 S. Liao, M. W. “Dominant Local Binary Patterns for Fig.5. Examples of false detecting (True negative) Texture Classification,” Transactions on Image Processing, Vol 18, 2009, No 5 14 T. Celik “Fast and efficient method for fire detection using image processing,” ETRI journal, 2010, №6, pp. 881–890. 15 T. Celik, H. Ozkaramanli, H. Demirel “Fire pixel classification using fuzzy logic and statistical color model,” Acoustics, Speech and Signal Processing, 2007, pp. 1205– 1208. 16 H. Tian, L. Wanqing, P. Ogunbona and others “Smoke Detection in Videos Using Non-Redundant Local Binary Pattern-Based Features,” Multimedia Signal Processing, 2010, pp. 1–4. 17 D. T. Nguyen, Z. Zong, P. Ogunbona, and W. Li, “Object detection using non-redundant local binary patterns,” in Proc. IEEE International Conference on Image Processing, 2010, pp. 4609–4612. 18 G. Zhao and M. Pietikä inen, "Dynamic Texture Recognition Using Volume Local Binary Patterns," Infotech Oulu and Department of Electrical and Information Engineering University of Oulu, Oulu, 2007. Fig.6. Examples of falsely detector work (False positive) Copyright © 2017 MECS I.J. Intelligent Systems and Applications, 2017, 2, 42-48 48 Video-based Flame Detection using LBP-based Descriptor: Influences of Classifiers Variety on Detection Efficiency 19 J. C. Burges. “A tutorial on support vector machines for How to cite this paper: Oleksii Maksymiv, Taras Rak, Dmytro pattern recognition,” Data Mining and Knowledge Peleshko,"Video-based Flame Detection using LBP-based Discovery, vol. 2(2), pp. 121-167, 1998. Descriptor: Influences of Classifiers Variety on Detection 20 U. Krebel “Pairwise classification and support vector Efficiency", International Journal of Intelligent Systems and Machines. In Advances in Kernel Methods: Support Applications(IJISA), Vol.9, No.2, pp.42-48, 2017. DOI: Vector Learnings,” MIT Press, Cambridge, MA, pp 255- 10.5815/ijisa.2017.02.06 268, 1999. 21 Jinho Kim, Byeong-soo Kim and Silvio Savarese, “Comparison Image Classification Methods: K-Nearest Neighbor and Support-Vector-Machines”, Proceedings of the 6th WSEAS International Conference on Circuits, Systems, Signal and Telecommunications”, Cambridge, USA, pp. 133-138, Jan. 25-27, 2012. 22 O. Maksymiv, T. Rak, O. Menshikova, "Deep convolutional network for detecting probable emergency situations", 2016 IEEE First International Conference on Data Stream Mining & Processing (DSMP), Lviv, Ukraine, pp. 199-202, 2016. 23 M. Pietikä inen, A. Hadid, G. Zhao and T. Ahonen, “Computer Vision Using Local Binary Patterns,” Oulu: Springer, 2011 24 B. Scholkopf and A. J. Smola. “Learning with Kernels,” MIT Press, 2002. 25 Ye zhiwei, Yang Juan, Zhang Xu, Hu Zhengbing,"Remote Sensing Textual Image Classification based on Ensemble Learning", International Journal of Image, Graphics and Signal Processing(IJIGSP), Vol.8, No.12, pp.21-29, 2016.DOI: 10.5815/ijigsp.2016.12.03. Authors’ Profiles Oleksii Maksymiv was born in Drohobych, Ukraine (1992), now he lives in Lviv, Ukraine. He obtained B.Sc. (2013) in fire safety and M.Sc. (2014) in project management. Currently, he is a Ph.D. student in computer science at the Lviv State University of Life Safety. His research interests focus on the computer vision, intelligent systems, image analysis and processing. Rak Taras, Dr. Sc., an Associate Professor and vice-rector on scientific and research work at Lviv State University of Life Safety. He received a Doctor of Sciences degree in Information Technology. He has published more than 100 papers in international and national scientific issues. The main research interests include information technology, decision-making system, control systems for emergencies. Dmytro Peleshko, Dr. Sc., Professor at Lviv Polytechnic National University, Lviv, Ukraine. He has published more than 100 papers in international and national scientific issues and journals and he is the author of several monographs. His scientific interests are image and video processing for the system of artificial intelligence Copyright © 2017 MECS I.J. Intelligent Systems and Applications, 2017, 2, 42-48

Advise: Why You Wasting Money in Costly SEO Tools, Use World's Best Free SEO Tool Ubersuggest.