Analysis of methods for reducing the number of false alarms in video-based fire detection systems

Abstract

The article focuses on a review and analysis of methods for reducing false alarms in video-based fire detection systems (VBFDS). The author of the article has designed a neural network and video-based flame detection algorithm to evaluate the effectiveness of methods found in the literature and other sources. The video-based flame detection algorithm was designed using a CIFAR-10-NET convolutional neural network. The D-Fire database, which contains 50000 fire images, was used to learn and test the algorithm. An error matrix was used to determine the effectiveness of the algorithm and methods to reduce the number of false alarms in video-based fire detection systems to determine parameters such as sensitivity (True Positive Rate, TPR), precision (Positive Predictive Value, PPV) and accuracy (ACC).