<p>To realize accurate and fast fire danger tracking in complex application scenes, a fusion method with aid of image gray value difference technique and Moth-flame optimization algorithm is developed. The method has two advantages. First, with this method, the critical image gray value difference can be employed to capture the dynamic characteristics of flames, thereby enhancing accuracy. It can overcome the interferences of complex fire scenes with many similar flame colors and provide a stable video-based fire tracking result. The method can successfully predict all our test cases with 100% prediction precision. Second, the method does not need prior data training. It can achieve fire danger tracking solely by relying on the currently monitored frame image within 11&#xa0;s. These advantages make the method significantly superior to traditional data-driven methods in terms of calculation accuracy and efficiency, especially for small flames whose area occupies less than 0.1% of the image. Traditional data-driven methods usually require hours of training and exhibit a low detection rate in such scenarios. Three numerical cases are implemented to validate the method. The results indicate that the method can successfully distinguish the fire danger zone from areas with similar flame colors in the monitored video, and even small flames in the early stage of a fire can also be tracked based on the method. Moreover, when compared with previous fire detection methods, the proposed method demonstrates significantly higher fire tracking accuracy. This advancement offers a critical technical solution for achieving fast and precise fire tracking in real-world applications.</p>