Fire Detection Based On Image Feature Extraction

Fire has always been a major natural diaster in the process of human civilization and development. It threatens our everyday lives such as in production, residential and puts us in grave danger. Therefore, automatic fire detection becomes a very important subject in our directions of studies. Our society is growing at an exponential rate and video surveillance is very commonly used in every aspect of life. Based on these factors, vision-based fire detection in videos has the potential to become a popular method in detection which is highly economical and applicable.In this paper, we propose a new method in identifying fire in videos. First, we analyze the characteristics of colours of the fire regions and extract the potential fire regions based on these characteristics. Then, we analyze its features such as motion, shape, surface coarseness and skewness because they are very commonly observed in fire; achieve automatic feature extraction; apply Fisher’s linear discriminant method to analyze the effectiveness of features in detecting fire in order to reduce its dimensions in features. Finally, we succeed in vision-based fire detection in videos by using the Bayes classifiers and evaluate the results.In the paper, we employed different backgrounds in videos in detecting fire. The experiments showed that the proposed method in vision-based fire detection in videos was effective; the results were accurate and can be used in real-time analysis.