Fire detection in Surveillance Videos using a combination with PCA and CNN

Abstract

This paper proposes a novel approach to early fire detection system from closed-circuit television (CCTV) using combination Principal Component Analysis (PCA) and Convolutional Neural Networks (CNN). It takes full advantage of the existing traditional methods like color or motional characteristics information of fire. However, CNN based fire detection system needs more computational requirements, high memory and time, in this paper, we propose energy-friendly CNN architecture for fire detection deep neural networks, inspired by MobileNet. The main role of PCA is to perform feature extraction of row data and then send it to CNN architecture. The experimental results on benchmark fire datasets reveal that the proposed method can achieve better classification performance and indicates that using CNN to detect fire in video captures is an effective way.