Real-time identification of smoke images by clustering motions on a fractal curve with a temporal embedding method

Automatic forest fire detection with CCD cameras requires a landscape image analysis in two stages: first the tracking of local dynamic envelopes of pixels, and second the discrimination between the various natural phenomena that may cause such envelopes. For this second process, we have to deal with restrictive conditions: lack of spatial information, complexity of motions, and real-time constraints on detection. We present here a fast algorithm adapted to the extraction of complex motions in small spatial envelopes, The principle of the method is to extract local motions from cluster analysis of points in a multidimensional temporal embedding space. We detail the four successive steps of this method: temporal embedding of gray-levels, fractal indexing of points, chaining points into a linked list, and motion extraction from point sequences of the linked list. (C) 2001 Society of Photo-Optical Instrumentation Engineers.