The Night Flame Detection Algorithm Based on Sequential Frame Difference

Flame detection has significance to reduce the loss of life and property caused by fire. At present, there is a lack of outdoor night flame detection methods. Most of the existing flame detection methods are based on flame characteristics or models, and most of them require additional storage and computation, which will reduce the system performance. The difficult problem of outdoor night flame recognition is how to eliminate the interference of fixed and mobile light sources. The paper proposes a method to detect outdoor night flame by using flame brightness and location characteristics. This method can detect the night flame accurately without additional cost. The frame difference method and OTSU algorithm are used to extract and segment the suspected flame targets in three consecutive frames. It can effectively reduce the interference of fixed and mobile light sources. We calculate the position coincidence rate of the suspected flame, and then compare the coincidence rate with the preset threshold to determine whether the suspicious target is a flame. The effectiveness of our proposed method is validated by experiments carried out on our self-created dataset, which achieves 95.5% detection accuracy.