Intelligent Surveillance Systems for Fire Detection in Open Areas: A Survey

Abstract

With the growth of open areas comes an ever-increasing risk of fire. However, there is a problem with the present approaches to fire detection, which rely on smoke sensors for wide regions. The advent of video surveillance systems has greatly improved our ability to detect smoke and flames coming from a distance and reduced this risk. Point sensors are slower at detecting fires than cameras when image processing is used. Moreover, using this video and image data presents processing challenges due to the enormous volume of data involved. Several approaches have recently been put forth to address this issue and distinguish between fire and smoke. Earlier methods included image processing algorithms for flame and smoke detection as well as motion-based estimation of smoke. Recently, a variety of techniques have been put forth using deep learning and convolutional neural networks (CNNs) to predict and automatically identify fire and smoke in videos and images. In this study, we review previous studies of fire/smoke detection systems based on machine vision and deep learning. The foundations of image processing techniques, CNN, and their applicability to video smoke and fire detection are explained. A discussion of current data sets and an overview of recent methods applied in this field The difficulties and potential solutions for advancing the application of CNN in this field are discussed. Then a comparison for researchers in the last years based on the dataset, year, technique, limitation, and accuracy they got The CNNs have been found to have a high potential for detecting open areas, and improved development can help create a system that would significantly reduce the loss of human life and property. Remarks for future work were concluded.