Analysis of Deep Learning Techniques Used for Indoor Flame Detection

In spite of the Government placing multiple norms, fire remains one of the major causes of disaster. Fire stations being at a 65% shortfall, precaution becomes our only choice. While the traditional models relied on hand-crafted features, the new Deep Learning techniques have made it possible to extract higher-order features through extensive training. This article enlists and examine the various deep learning architectures such as CNN, R-CNN, Faster CN, YOLO, SIFT, etc. employed for computer vision-based flame recognition. A detailed analysis will lead to a better understanding and thus help the researchers to identify problem statement and design improved solutions. The article also includes the different datasets referred till date and the common metrics used for measuring the performance of any given model. A single point assimilation of all the aspects of designing an algorithm for efficient flame detection, starting from understanding the set of features unique to flame images, a robust model, a diverse dataset and finally a set of reliable metrics will encourage easy identification of gaps so that effective solutions may be proposed in this domain.