Fire And Smoke Detector

Abstract

Fire incidents are very dangerous and devastating. They happen every year which ends up in great losses. Annually, 5 to 8% of the 3.3 million premature deaths are due to fire emissions and are increasing in number every year. Fire and smoke systems are installed to detect the fire and smoke so that preventive measure could be taken. It has been witnessed that fire and smoke spread very fast with irregular patterns which causes massive destruction in no time. Due to this reason, it has been observed that current detection systems are not able to detect the fire and smoke earlier. Therefore, better detection system is needed to detect fire and smoke. This study presents a Machine Learning based real-time video fire and smoke detection system that can detect fire and smoke through camera and notify authorities at multi-level according to the severity of fire. For the development of the system benchmark dataset is collected which contains 1500 images. Dataset consists of images having different severity of fire, smoke and non fire/non-smoke which are classify according to the requirement of the system as their names. This dataset is used to train the model. A pre-trained model of VGG-16 is used for the development of this system. Design of the model is based on VGG-16 a pre trained Convolutional Neural Networks (CNN) model along with stack of different layers to improve the accuracy. Method of hyper-parameter tuning is used to analyse the behaviour of the discussed model. Model is tested and trained for different range of values of Batch size, learning rate and epochs. Performance of the model is validated through 5-fold cross validation. Experiments and analysis show that model detects the fire and smoke with 98% cross validation accuracy. Fire and Smoke Detection System (FSDS) is developed as a desktop application for ease of use. In this system trained model is imported at the backend of the application which is used to detect the fire and smoke. System detects fire or smoke by extracting frames from live feeds through camera. Moreover, system also notifies the authorities at multi-level (three levels) according to the severity of fire by sending fire alerts to take safety measures as early as possible. In this age where computers can now perceive and analyse their environment with high accuracy, the detection system use latest and optimized model which specializes in object detection