Early smoke detection in forest areas from DCT based compressed video

Smart functions begin to be integrated into some camera acquisition systems for CCTV applications. In this regard, smoke detection based on the compressed video is a highly desirable functionality for the monitoring of forest that present a high risk of fire. In this paper, we propose a fast and early smoke detection method that measures the local fractal feature of smoke areas based on the Discrete Cosine Transform (DCT) coefficients. The latter are generally computed at the compression step which is concomitant in its acquisition. The two coding video standards MJPEG and MPEG2 are considered as they are widely available in the operating cameras. The novelty of our approach consists in resorting to a recursive DCT in order to improve the detection performance.