Automation of fire watcher : sentry approach

Abstract

Residential fire is one of the most neglected problem in Singapore. Very few residential premises are installed with fire detection system. The situation is becoming worse as more and more people are getting e-scooter at home. These e-scooters could easily catch fire or even explode due to over-charging. This causes the residential fire more frequent and uncontrollable. Most of the fire detectors are not effective in implementation in residential premises due to false alarm. Cooking or burning incense paper could easily trigger the fire detector. These activities are, in fact, not a real fire situation which generally happen in every household on a daily basis. Without any better solution specifically for residential application, excessive false alarm will happen and cause unnecessary disruption to the residents. This project aims to propose a new fire detection framework that could be effectively implemented in residential premises. We’ve come out with an idea of using artificial intelligence to detect the elements of SAFE state of fire. This idea is mainly inspired by industrial fire watcher and ‘no-fire detector’ by Rainer Siebel. 4 hot works (activities with elevated temperature to achieve useful purpose) that generally happen in residential premises are targeted in this project, including cooking, burning incense paper, lighted cigarette and mosquito coil. The results show that our approach succeed to identify the hot works and being more immune to false alarm as compared to smoke detector. Our approach could complement with the existing detector to work as a 2-layer detection system, stopping the detector from triggering unwanted alarm if the hot work is detected by our framework.