Kovvuri Narendra, et. al. International Journal of Engineering Research and Applications www.ijera.com ISSN: 2248-9622, Vol. 10, Issue 11, (Series-II) November 2020, pp. 01-07 www.ijera.com DOI: 10.9790/9622-1011020107 1 | Page A New Approach to Vision-Based Fire Detection Using Machine Learning 1 Kovvuri Narendra, 2 Dr. Jhansi Rani Singothu 1 M.tech in CST With Artificial Intelligence And Robotics, Department of Computer Science And Systems Engineering, Andhra University College Of Engineering(A),Andhra University ,Visakhapatnam, AP, India 2 Assistant Professor, Department of Computer Science And Systems Engineering, Andhra University College Of Engineering(A),Andhra University, Visakhapatnam, AP, India. ABSTRACT Computer vision - based fire detection has recently attracted great deal of attention from the research community. In this paper, the authors propose and analyses a new approach for identifying fire in videos. Computer vision techniques are largely used now a days to detect the fire. There are also many challenges in judging whether the region detected as fire is actually a fire this is perhaps mainly because the color of fire can range from red yellow to almost white. So fire region cannot be detected only by a single feature color many other features have to be taken into consideration. This paper is a study of the recent techniques and features extracted by different existing algorithms. In this approach, we propose a combined algorithm for detecting the fire in videos based on the changes of the statistical features in the fire regions between different frames. The statistical features consist of the average of the red, green and blue channel, the coarseness and the skewness of the red channel distribution. These features are evaluated, and then classified by Bayes classifier, and the final result is defined as fire-alarm rate for each frame. Experimental results demonstrate the effectiveness and robustness of the proposed method. There is different method which focuses on various properties of fire like, color, shape, movement, spatio-temporal features etc. For real-time identification of fire from videos simple and accurate method is proposed as multi expert system, which uses color, shape and movement evaluation for detecting fire. The study refers different methods for fire detection and prefers integration of smoke analysis for early identification of fire. KEYWORDS: Fire detection, Pattern recognition, Bayes classification, Flame feature, Probabilistic model, statistical Features of Fire Regions --------------------------------------------------------------------------------------------------------------------------------------- Date of Submission: 31-10-2020 Date of Acceptance: 12-11-2020 --------------------------------------------------------------------------------------------------------------------------------------- I. INTRODUCTION Two main applications of vision-based fire detection are: (1) monitoring fires and burn disasters from surveillance systems [1], and (2) automated retrieval of events in newscast videos [2]. These applications play an important role in modern society. Recently, there have been a number of efficient methods proposed for vision-based fire detection in[1]-[6].In [1], Chao-Ching Ho analyzed the spectral, spatial and temporal characteristics of the flame and smoke regions in the image sequences. Then, the continuously adaptive mean shift vision tracking algorithm was employed to provide feedback of the flame and smoke real-time position at a high frame rate. P. V. K.Borges and E. Izquierdo, in [2], analyzed the frame-to-frame changes of specific low-level features such as color, area size, surface coarseness, boundary roughness, and skewness within estimated fire regions to describe potential fire regions and used Bayes classifier to indicate a frame contains fire or not. In [3], CelikT. et al. developed two models, one for fire detection and the other for smoke detection. For fire detection, the concepts from fuzzy logic were used to make the classification fire and fire-like colored objects. For smoke detection, a statistical analysis was carried out using the idea that the smoke shows grayish color with different illumination. In [4], the authors also used a probabilistic metric to threshold potential fire pixels Since the fire causes serious damages, fire detection has been an important study to protect human life and surroundings. As the economy develops, number of large buildings also increases. If fire happens in these buildings then there will be a bad social impact, major property damage and heavy casualties will be easily caused. So fire should get detected early for extinguish and evaluation. In large buildings, rooms and outdoor places, fire detectors can hardly detect fire characteristic parameters like temperature, vapor RESEARCH ARTICLE OPEN ACCESS