Research On The Video Smoke Detection For Fire Alarm

With the development of society and economy, people pay more and more attention on the fire disaster in large public places and forests. Many researchers are working on the early detection of fire. Currently, by using of sensor technology, there are some fire detection equipments which are mainly rely on the perception of flame, temperature and smoke. Under this traditional technology, every sensor point could only monitor the limit space that around itself, and it is difficult to use these sensors to monitor fire in large spaces or high-speed airflow scenes. With the development of computer technology and digital image theory, researchers move their focus from the physical characteristics of pre-fire to the nature of fire, and the video fire detection is proposed. The video fire detection system makes use of digital image processing technology and pattern recognition technology, it could solve the problem of fire detection in large spaces by monitoring the image characteristics of fire and smoke. We usually see lots of smoke before fire disaster take place. Therefore, the smoke monitoring is important to the detection of fire.The technology of video smoke detection are studied exploratory in this paper, and have carried on the simulation experiment. The study mainly includes the following aspects:(1) The paper introduces the application of digital image processing technology in video smoke detection, and adopts a smoke motion region detection algorithm that based on background image update.(2) The paper proposes a flutter analysis based smoke feature extraction method. Besides, the density distribution feature and the boundary shape feature of smoke region are analyzed, and the extraction methods are also proposed. Finally, a 9-dimensional feature vector is used to describe the characteristics of smoke image.(3) The paper introduces the principle of ANFIS(adaptive neuro-fuzzy inference system), and its application in video smoke detection. Combined with the smoke feature vector, a structure and algorithm of video smoke detection system are presented. The experimental results show that the ANFIS algorithm has excellent performance on ROC(receiver operating characteristic) curve which is very important for real application.(4) After the introduction of bayesian decision, a video smoke detection algorithm that integrates with bayesian decision is proposed. The bayesian decision is presented at the output of ANFIS, and the minimum error-based bayesian decision and the minimum risk-based bayesian decision are both used to fix decision result. The experimental results show that the minimum risk-based bayesian decision integrated algorithm could achieve better performance than primitive ANFIS algorithm.