DEVELOPMENT OF COMPUTER VISION-BASED AUTOMATIC FLAME DETECTION ALGORITHM USING MATLAB SOFTWARE ENVIRONMENT

. Fire detection systems plays an important role in protecting objects from fires and saving lives. In traditional fire detection systems, fire detectors detect fires by combustion of by-products, such as smoke, temperature, flame radiation. This principle is effective, but unfortunately, the fire detector works with a significant delay if the ignition source is not in close proximity to it. In addition, such systems have a high frequency of false positives. The most promising area for early fire detection is the use of computer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systemscomputer vision based fire detection systems, as they detect fires rather than their combustion products. Such systems, as well as traditional fire detection systems, analyze the signs of a fire, such as smoke, flames, and even the air temperature by means of the image coming directly from the cameras, due to which the range of the system increases significantly. Unlike traditional systems, they are more efficient, do not require indoor spaces, have high performance and minimize the number of false positives. In addition, when notifying the operator about a fire, the video system can provide him with an image of probable ignition place.Fire detection algorithms are quite complex because the signs of a fire are non-static. Today, more and more scientists are trying to develop algorithms and methods that will detect fires at an early stage in the video stream with high accuracy, without false positives. When creating such algorithms, there are four main approaches. These are flame colour segmentation, motion de-tection in the image, analysis of spatial changes in brightness and analysis of temporal changes in boundaries. Each approach requires the development of its own individual algorithm, combining them, which is quite a difficult task. However, all algorithms are based on the process of selecting colours in the image that are characteristic of fire. There are many algorithms that use two or three approaches and they provide good results. Using the MATLAB software environment and its standard packages to create a flame detection system is considered in this paper.Purpose. The research aims to develop an algorithm for automatic flame detection in images based on pixel analysis, which identifies the colour of the flame and flame area using the MATLAB software environment, in order to further create a reliable computer vision-based flame detection system.Results. The MATLAB software environment includes Image Acquisition Toolbox and Image Processing Toolbox, which are compatible environments for developing real-time imaging applications that can come from digital video cameras, satellite and aviation on-board sensors, and other scientific devices. Using them, one can implement new ideas, including the development of fire detection algorithms.The flame has a fairly uniform intensity compared to other intensities of objects, unlike smoke. That's why there are so many flame-based fire detection algorithms. However, in practice, developing an effective algorithm is not an easy task, because the image under study may contain objects that have signs of flame. In the image, you need to select the pixels with the characteristic colour that are inherent in the flame. At this stage, various images with flames in the RGB colour model were analyzed and the mean value of their intensity and standard deviation (R, G and B) were determined. Image segmentation was performed on the basis of the obtained values. The purpose of segmentation was to highlight the flame in the image. However, there may be other objects in the image whose pixel intensities match the flame pixel intensities. As a result, in addition to the flame, other objects may be highlighted in the segmented image. Based on the previously selected segmentation method, we can assume that the flame in this image occupies the largest area. Therefore, another criterion was chosen for the flame search, based on the area, which enabled to remove other objects that do not belong to the flame. In the final stage, the flame in the image is highlighted by a rectangle.Conclusions. The possibility of using the MATLAB software environment with the Image Acquisition Toolbox and Image Processing Toolbox packages to create a computer vision based flame detection system is considered. The functions of the packages allow you to implement new ideas when creating algorithms for automatic fire detection. The article develops the algorithm for automatic flame detection in the image based on the analysis of flame colour pixels and flame area. Various images with flames in the RGB colour model were analyzed and their mean value and standard deviation were determined. Image segmentation was performed on the basis of the obtained values. Experimental studies in the MATLAB software environment have proved the effi-ciency of the developed algorithm. To create a reliable computer vision based flame detection system in future, it is proposed to develop an algorithm that would analyze the boundaries, shape and flicker of the flame in addition to analyzing the flame colour pixels and flame area.