Fire Detection System based on integrated Infrared-visible Cameras

An integrated information system developed for the monitoring and the automatic detection and location of forest fires is described. This system uses robotized stations equipped with combined infrared (IR) and visible cameras. Different models of integrated visible and IR acquisition systems have been developed based on computing thermal and spatial information suitably fused . Recognizing a fire using only IR cameras is subjected to problems due to both the transparency of the smoke to Far-IR wavelength (i.e. 8?12 ???m), and the possible masking of the interested area. The use of visible images related to the same area, improves the possibility of image processing, in order to identify and locate fire areas as earlier as possible after the fire has started. The developed system includes two coupled, aligned, and mobile cameras, one acquiring at Far-IR wavelengths and the other at the visible spectrum wavelengths. Series of multiple images are acquired, spanning all the area under surveillance, from both the cameras. A comparison of the actual radiance gradient values with reference values acquired a priori, gives a first discrimination for possible fire location. Then the visible images can be compared to other pre-acquired reference images, analysing only fragments in search of identifying elements to give evidence to the fire location previously detected. The analysis of the IR images includes also a filtering based on both dimension and eventual movement of the individuated target. Furthermore, the temporal oscillations of the possible fire source are examined in order to discriminate between light reflection (which are more constant) and fires. In order to increase the performance of the system, decreasing the rate of false positives, additional information about the terrain usage, including a fire risk factor, a digital terrain model (DTM), meteorological and other information, is previously stored in a database and then used to compute the final output which can result in an alarm or not . A set of fuzzy rules defined a priori are then used, in the context of a decisional component, to yield a final global rate regarding the possibility of a fire in the previously localized area. The system has been successfully tested for forest fire monitoring in S. Rossore Natural Park near Pisa