Patch-wise periodical re-occurrence analysis of motion for real-time video fire detection

In this paper, an approach for video fire detection is proposed. The basic idea is fire has a highly varying texture and waits at the same location for long consecutive periods. Since pixel-wise framework has high algorithmic cost, a patch-wise periodical analysis of fire colored moving pixels in terms of consecutive re-occurrence in a given duration is presented. Instead of analyzing the fire motion frame by frame, motion ratio of each patch is evaluated at the end of each period to attain true turbulence of fire. Experiments with a large number of fire and non-fire video sequences give promising results.