Flame detection algorithms based on temporal-spatial visual saliency

Introduced the idea of visual saliency in computer vision, the model of flame detection is proposed based on the temporal-spatial visual saliency. Firstly, the Lab color space is obtained, then three channels (L, a, b) are segmented with threshold and filtered in Gaussian. The grayscale integral projection method is used to extract the brightness saliency of flame. Secondly, the flame region of interest (FROI) is acquired by frame difference of flame color, and the region is determined through dispersion and cumulative movement. Finally, the integral saliency map of the current frame is formed by compositing color, brightness and motion saliency. Experiments show that the proposed model achieves better performance on flame detection than others under different video scenes.