Automatic Fire Smoke Detection Based on Image Visual Features

Abstract

For open spaces, this paper proposes a novel method for automatic fire smoke detection based on image visual features. The greatest characteristic of the method is that both static and dynamic features of fire smoke are investigated. And the basic strategy is that we extract features of the moving target including growth, disorder, frequent flicker in boundaries, self- similarity and local wavelet energy as a joint feature vector which will be normalized, and then a BP artificial neural network is trained to recognize fire smoke. Experimental results show that this method can achieve early detection of fire accident with high accuracy and stronger anti-jamming ability.