Skip to Main Content
In this study, we propose a novel color context analysis based efficient real-time flame detection algorithm (CCAFDA). To measure the relevance of color context of every two adjacent flames in flame image sequences, two new flame feature vectors are defined: one is the flame detection context based dynamic feature row vector and the other is the optimal flame feature area vector. The proposed algorithm uses the flame detection context based dynamic feature row vector and the optimal flame feature area vector as the joining point between every two adjacent flames and then according to the relationship of color context of multiple frames as well as the relevance between the adjacent pixels, selects the area of optimal flame feature in real time and adjusts the area of optimal flame feature dynamic. The proposed methods only scans the optimal flame feature area in each frame rather than scans every single pixel for each frame, and then uses the burning degree of the optimal flame feature area as a measurement to estimate the burning degree of the whole fire flames. To compare with the conventional method which scans the whole flame video through point by point scanning in RGB color space, the proposed methods improved the efficiency which can detect the flame status in video stream in real time. Experiments show the proposed algorithm improved the efficiency for detection and estimate of the boiler flame.