An image processing technique for automatically detecting forest fire

In this paper we present an automatic system for early smoke source detection through the real time processing of landscape images. The first part describes the segmentation technique we use to extract persistent dynamical envelopes of pixels into the images. We describe the temporal algorithm at the pixel level (filtering) and the spatial analysis to bring together connected pixels into the same envelopes (object labeling). The second part deals with the method we use to discriminate the various natural phenomena that may cause such envelopes. We describe the image sequence analysis we developed to discriminate distant smokes from other phenomena, by extracting the transitory and complex motions into little pre-processed envelopes. We present then our main criterion for smoke recognition based on the analysis of the smoke plumes velocity. (C) 2002 Editions scientifiques et medicales Elsevier SAS. All rights reserved.