Real-time fire detection from image sequences is a crucial requirement in video surveillance applications, enabling the prevention of environmental disasters and ensuring continuous monitoring of urban and forest areas. To meet this demand, there is a growing interest in deploying "intelligent cameras" equipped with onboard video-analytics algorithms capable of detecting fires, such as flames and smoke, in real-time. These cameras are typically deployed in remote locations and need to be self-sufficient in terms of computational resources, with a small embedded system capable of processing the image sequence using a fire detection algorithm. In this context, our work presents a novel network model called FireNetAGM, which leverages the MobileNetV2 architecture for feature extraction from a sequence of video frames and utilizes an LSTM architecture to handle the temporal dynamics of the video.
Gen-2.CCTV.CAMERAS.DETECTION.mp4
