A novel fire recognition algorithm based on flame´s Multi-features Fusion

In this paper, a novel fire recognition algorithm based on muti-features fusion(MFF) with self-adaptive, self-learning and fault tolerance is proposed after analyzing the characteristics of the burning fire, and taking the flame´s twinkling frequency and its dynamic contour into account. First, picking up the dynamic video of fire, extracting the eigenvector of suspicious fire area, and using disperse FFT to the eigenvector to build a quantified distinguishing rule; then, checking the nearby frame which changes most, qualitative and quantitative analyzing its changes to form a quantified edge feature. After quantifying all the characteristic parameters, a probability update function is used to fusion such muti-features to carry out the probability recognition of fire. Tests show that MFF algorithm of this paper has the abilities of self-adaptive, self-learning and anti-interference to white spots from moving or static objects, even under a very complex circumstance with spotlight, automobile light and illumination changing.