Mixed Approach of Real-Time Smoke and Fire Recognition from CCTV Cameras

Abstract

To date, the main problem associated with the prevention of fires in open areas is their early detection. It is very difficult to put out a fire that has already begun, even with large modern firefighting equipment and human resources. However, using modern artificial intelligence technologies, it is possible to offer more effective ways to solve the problem - this is the detection of smoke and fire at the initial stages of ignition. The most promising method for early detection of a fire is its detection based on video surveillance. Monitoring of large areas and the rapid decline in the cost of video cameras will allow this latest technology to become truly widespread in the fire safety system. This article proposes object recognition by hybrid methods based on contour analysis and neural network technologies. Tests of the results of smoke and fire recognition by the methods of contour analysis and artificial neural network were carried out by the full-scale method in accordance with the test method defined by GOST R 53325-2012.