Fire detection for early fire alarm based on optical flow video processing

We propose a method for early fire detection based on the Lucas-Kanade optical flow algorithm that is able to detect fire in real time in a video stream from a monocular camera. The method works both indoors and in open areas. It detects fire at the beginning of the burning process, enabling an earlier response than would be possible with a conventional fire detector. The method performs background subtraction to identify moving pixels in the scene then filters for colors consistent with fire. Growth rate analysis is applied to the contours identified in the previous step. We further propose the use of a Lucas-Kanade optical flow pyramid in the regions identified in the growth analysis step and analysis of the variance of the optical flow feature points' motion. In an experiment comparing the growth rate method alone with the combination of growth rate and optical flow analysis, we find that the algorithm improves the accuracy of fire detection for early fire alarming and reduce false alarm by solving the fire-like color.