Real-time fire detection in camera stream using statistical analysis

Abstract

The paper describes a new algorithm designed to be fast and efficient for detecting fire. It is based on finding and investigating suspicious regions in each frame of video stream. The investigation consists of tracking regions across frames and performing statistical analysis on their trajectory. If the trajectory has characteristic similar to fire, a test on persistence is performed. If the fire-like characteristics persists, the alarm is triggered. This criterion enables to eliminate a large proportion of false alarms. Given it’s simplicity, the algorithm can be used separately in some environments and can improve existing algorithms as well. © 2017, World Scientific and Engineering Academy and Society. All rights reserved.