Fire Detection Through Surveillance Videos Using Deep Learning in Real-World Applications

Abstract

Unearthing fire and smoke in sighted landscapes is challenging because of the great range of texture and color. To address this issue, several fire and smoke picture taxonomy systems have been proposed; however, the majority of them depend on either rule-based approaches or hand-crafted attributes. The technologies that support fire and smoke detection systems are important to ensuring and providing maximum performance in today’s surveillance situations. To overcome these limits, describe a unique technique based on a deep learning approach that employs a convolutional neural network with expanded convolutions. Fire may cause significant loss of life and property. Evaluated our method by training and testing it with a custom-built dataset of fire and smoke photos obtained from the Internet and manually classified. In terms of performance, our approach was compared against solutions based on well-known snipping architectures. The CNN used to recognize the activity of video abnormality and if any fire or smoke image was captured while analyzing frame by frame video then it was stored in the firebase database. If any fire frame is captured in the dataset then it gives the alert message to the android like notification. In terms of classification performance and complexity, our results suggest that our approach outperforms others.