A Survey of Deep Learning Methods for Vision-Based Fire Detection and Localization

Abstract

Undetected fires have caused significant damage globally, highlighting the importance of a reliable fire detection system. However, current smoke and fire detectors have proven to be inadequate due to system inefficiencies. To address this issue, a surveillance camera-based vision-based fire detection system with a high detection rate and low fault warning rate has been proposed. This system utilizes live video data analysis for real-time fire detection and incorporates edge detection and thresholding approaches to recognize fire attributes. The model identifies harmful flames based on their color, velocity, form, and texture, and employs color models such as HSV to increase detection accuracy. Deep learning methods like YOLO and transfer learning, as well as IoT technologies, can be incorporated to improve fire detection both indoors and outdoors. This paper presents a comprehensive literature review on various approaches for the detection and localization of fires in indoor and outdoor environments. The paper highlights the strengths and limitations of each approach. It also discusses the challenges and future research directions in this field to identify opportunities for advancing the state-of-the-art in vision-based fire detection and localization.