Wildfire Detection in Video Images Using Deep Learning and HMM for Early Fire Notification System

摘要

Conflagration is disaster that strongly influence environmental ecology and lives. In building, sensors are sufficient to support notification and evacuation. In contrast, the outdoor fire notification is supported via image fire detection. For wildfire notification and assistance in rescue, many researches extract fire characteristic from digital image to detect fire. However, the false alarm of detection system happens when it find the little flame can be extinguished itself with time. In case of false alarm often happens, the rescuer may neglect it and lose the first opportunity to rescue in mission. In this paper, we propose a system that combine deep learning and hidden markov model (HMM) in order to reduce frequency of false alarm. Our proposed system defines the status transformation of images/frames. Afterwards, the relationship of status transition is employed to notification classes. As experimental result, the fire detection rate of all deep learning architectures can achieve more than 96%. And our proposed system can also reduce up to 88.54% false alarm rate to decrease the waste of emergency manpower effectively.