Fire Video Image Detection Based on a Convolutional Neural Network

Abstract

Abstract Computer vision-based and real-time flame detection is very important to inmodern surveillance system. At present, convolutional neural network (CNN) has become a topic discussed by more and more researchers because of its high recognition accuracy and wide application. The preprocessind process of the traditional image processing method is complicated and the false positive rate is high. So, in this paper we proposed an algorithm for detecting flame in real time using CNN technology. Firstly, to improve the accuracy of detection, we proposed a suspicious target regions segmentation for disposing the suspected flame regions. This algorithm could locate the target area and segment the target area to improve the flame detection and recognition accuracy. Then, we designed a model based on CNN to classify the extracted feature maps of candidate areas. Finally, we could get the detection of flame according to the the classification results. The experimental results show that the approach has high recognition accuracy.