Nonparametric membership functions and fuzzy logic for vision sensor-based flame detection

This paper proposes an advanced fire-flame detection algorithm using camera images for a better performance than conventional sensor-based systems that are limited to a small area. First, candidate flame regions are detected from the captured images using a background model and flame-color model. After forming probability density functions for the intensity variation, wavelet energy, and motion orientation on a time axis, these probability density functions are changed into membership functions for fuzzy logic. Finally, the result function is made by defuzzification, and the probability value of a fire flame is estimated. The proposed algorithm is successfully applied to various fire videos, including indoor and outdoor fires, and shows a better detection performance when compared with other methods.