Fire Detection System Using Color and Flickering Behaviour of Fire with Kekre's LUV Color Space

Abstract

This paper presents a vision based fire detection system based on Kekre's LUV color space for flame detection and uses the flickering nature of fire to confirm the existence of fire. Traditional fire detection system based on sensors needs to be positioned at certain distance and works on certain trigger condition. The popularity and use of video surveillance at various places has facilitated the popularity of vision based fire detection system in both indoor and outdoor locations. The most important feature of fire is the flame color which varies from red, orange, yellow to white. In order to improve the accuracy of fire detection system, fire like non fire object needs to be differentiated from actual fire objects. The proposed technique uses Kekre's LUV color space to detect the presence of fire pixels. L gives luminance and U and V gives chromaticity values of color image. Negative value of U indicates prominence of red component in color image and negative value of V indicates prominence of green component over blue. Once the fire pixels are detected, the flickering nature of fire is used further to eliminate the fire like non fire objects and hence reducing the false alarm rate. Experimentation is performed on subset of MIVIA database with 21 fire videos is tested with RGB, HSV, YCbCr and Kekre's LUV color space to compare the accuracy and efficiency of the system. The accuracy of fire detection with RGB color space was 66.67%. The use of HSV and YCbCr color model provided an accuracy of 80%. Our method using Kekre's LUV color space for fire detection provided an overall average accuracy of 93.33%.