Image processing based deflagration detection using fuzzy logic classification

Abstract Image processing-based deflagration detection is currently a novel application with a considerable development potential. Moreover, this technology could replace the commonly used infrared photodiode-based detection sensors and can help to avoid tunnel disasters or accidents in chemical plants more reliably. Today׳s deflagration detection systems only provide a detection signal without any further information concerning the triggering event. In addition, these systems are not able to distinguish between a hazardous deflagration and a less dangerous fire-like process. This paper proposes a two-stage algorithm for deflagration detection in order to obtain this valuable information. The first stage identifies probable deflagration-like or fire-like pixels by their chromatic characteristics and dynamic intensity behaviour. The following stage evaluates the temporal expansion of the counted pixels using a defined spatial expansion parameter (SEP). In parallel to this, the oscillating change in the number of identified pixels over time is transformed into the frequency domain. The analysis of the frequency spectrum facilitates identifying fires by their typical flicker frequency. The proposed detection method uses fuzzy logic classification for each stage. Thereby no static thresholds are necessary, which yields more setting options in order to increase algorithm flexibility. Finally, the entire algorithm is tested in different realistic scenarios with focus on deflagrations. As a general result of the performance tests, the algorithm is able to detect and distinguish deflagrations and fires with high accuracy. Furthermore, the expansion of the detected combustion processes is described quantitatively.