FIRE DETECTION IN VIDEO SURVEILLANCE APPLICATIONS

Abstract

Fire is one among the most frequently occurring hazards and is the primary cause of disastrous individual injury and crushing property harm. Hence a framework for the early recognition of fire is important to keep fires from fanning out of control. Fire alarms are often employed in ordinary structures using sensors based upon physical signals, such as infrared flame detectors that release heat, smoke sensors and ultraviolet flame detectors, and so on. However, these necessitate significant human involvement, such as going to the scene of fire to verify there is a fire on receiving fire alarms. Hence, a two-stage fire detection technique utilizing Convolutional Neural Networks and Recurrent Neural Networks has been proposed. In the first stage, a pre-trained CNN model is used to extract the flame characteristics. In the second stage, feature vector from CNN reaches the RNN which then predicts the probability of the input being a fire or a non-fire event.