Towards microscope-video-based fire-detection

Nowadays, video-based fire-detection is a nearly common application, as the number of installed surveillance video-systems is growing and the related processing units getting more powerful. The fire-detection-feature is usually only an add-on feature of 'intelligent' surveillance video-systems. The detection-criteria are therefore based on macroscopic characteristics observable in the surveillance video, like particular smoke dynamics, flame flickering or loss of image contrast due to obscuration by the smoke. In a new approach not the macroscopic, but the microscopic characteristics of aerosols are analysed with view to a more reliable discrimination between fire and non-fire aerosols. By monitoring an illuminated sheet of air some centimeters under the ceiling of a room, it is possible to get information about a limited range of the particle size distribution, density and flow characteristics of the suspended aerosol. A prototype of the scanning system has been developed and a series of test-fires (according to the EN54) and non-fire tests has been carried out in the Duisburg-Fire-Detection-Laboratory. The comparison between the test-fires and the non-fire scenarios shows that the discrimination between fire and non-fire aerosols is cogitable. The analyses of the aerosol characteristics are based on adapted pattern recognition techniques applied in the image processing. Some differences between different aerosol types are even visible to the naked eye. This paper presents the developed prototype specifying important features. Afterwards the most interesting results are shown and commented. Finally, the possibilities and limitations of an automatic fire-detection system based on the microscope-video analysis are discussed.