Fire Recognition Based on Multiple Features of Video Images

Abstract

As traditional fire detection techniques have many disadvantages, the fire image automatic recognition algorithm based on video image processing is studied. After the images are preprocessed, multiple fire features will be extracted. Feature extraction has a great influence on fire recognition accuracy. According to video sequence features and the characteristic difference from the flame and other things in background, many flame features such as frame similarity, geometry, color, shape and other features are extracted. Grid searching method is utilized for support vector machine (SVM) to select better parameters. Then these features including frame similarity, red saturation value, area, eccentricity, circularity degree and other features of flame and other interferences are trained by SVM, and finally the fire images can be recognized. The experimental results show that the algorithm is effective, automatic fire detection has higher accuracy and decreases false alarm ratio.