Utilizing Convolutional Neural Networks (CNNs) for the Detection of Smoke and Fire through CCTV Camera Surveillance

Fires frequently cause large losses in terms of both property and human life. Thus, there is an urgent need for fire detection systems that are easily accessible, reasonably priced, accurate, and quick. The results of earlier attempts to create workable solutions have been inconsistent. A significant obstacle is striking a balance between model size and system performance, with the latter allowing portability to devices such as Raspberry Pi. This study proposes FireDetector, a neural network designed with performance and scalability in mind. FireDetector specifically seeks to (i) surpass current methods in accuracy and (ii) stay small enough for hardware deployment. Positive findings are obtained from evaluations using newly introduced bespoke fire datasets and benchmark datasets.