Abstract:
Traditional fire detection methods utilize sensors to collect information on smoke particles, flame temperature and relative humidity, and then, evaluate and respond to fires. Since these sensors must be placed near the flame, the traditional detection methods cannot be used in scenes with extreme interference to the sensors. Besides, these traditional methods are also difficultly applied in large spaces, open spaces, and complex scenes. Especially, these traditional methods difficultly confirm descriptive information such as fire location, flame size, and fire development status and have low real-time performance and accuracy. In order to prevent the occurrence of fires, the flame should be detected quickly and accurately. Video flame detection is suitable for complex physical environments, and has low cost, high detection rate, and short response time. By analyzing the static and dynamic characteristics of the flame, this paper proposes an Ohta color space detection and combination of saturation with Otsu threshold segmentation method, which can effectively outline the possible region of flames in real time. By circularity, rectangularity, center-of-gravity-height coefficient, combining with LBP texture analysis, these regions can be featured and judged via SVM method. Finally, it is shown from experimental results that the proposed algorithm can accurately detect the flame and effectively improve the real-time performance and accuracy.