Fire detection based on hidden markov models

International Journal of Control, Automation, and Systems (2010) 8(4):822-830 DOI 10.1007/s12555-010-0414-2

http://www.springer.com/12555

Fire Detection Based on Hidden Markov Models
Zhu Teng, Jeong-Hyun Kim, and Dong-Joong Kang* Abstract: In this paper, a novel method of real-time fire detection based on HMMs is presented. First, we present an analysis of fire characteristics thatprovides evidence supporting the use of HMMs to detect fire; second, we propose an algorithm for detecting candidate fire pixels that entails the detection of moving pixels, fire-color inspection, and pixels clustering. The main contribution of this paper is the establishment and application of a hidden Markov fire model by combining the state transition between fire and non-fire with fire motioninformation to reduce data redundancy. The final decision is based on this model which includes training and application; the training provides parameters for the HMM application. The experimental results show that the method provides both a high detection rate and a low false alarm rate. Furthermore, real-time detection has been effectively realized via the learned parameters of the HMM, since the mosttime-consuming components such as HMM training are performed off-line. Keywords: Fire detection, HMM, real-time processing, visual surveillance.

1. INTRODUCTION A fire detection system that enables accurate and prompt detection of fire is one of the most significant components of video surveillance and monitoring systems, since fire can cause loss of life and severe damage to property.Conventional methods generally utilize narrow range detectors such as chemical or gas sensors; however, these conventional sensors are suboptimal. First, the accuracy of a fire detection system closely depends on the precision and reliability of its sensors, the size of the sensing space, and the distribution of the sensors. If a higher precision fire detection system is needed, it is necessary todistribute the sensors densely in the detection space. Second, false detection can easily occur from other sources of smoke or fire, including even a lit cigarette. Third, there are fatal time delays for sensors to detect fire or smoke, which may result in the spread of fire, because the alarm is not issued until fire gas or chemical particles reach the sensors and activate them. Lastly, the expense ofsensors can impact the system cost. In contrast, video processing techniques offer many advantages. Video surveillance and monitoring systems have already been used in many

__________ Manuscript received May 4, 2009; revised January 26, 2010; accepted March 5, 2010. Recommended by Editorial Board member Jang Myung Lee under the direction of Editor Jae-Bok Song. This research was financiallysupported by the Ministry of Education, Science Technology (MEST) and Korea Institute for Advancement of Technology (KIAT) through the Human Resource Training Project for Regional Innovation, and partially supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (2009-0090165). Zhu Teng, Jeong-HyunKim, and Dong-Joong Kang are with the School of Mechanical Engineering, Pusan National University, 30 Jangjeon-dong, Geumjeong-gu, Busan 609-735, Korea (emails: {tengzhu, mare, djkang}@pusan.ac.kr). * Corresponding author. © ICROS, KIEE and Springer 2010

buildings and human environments due to rapid developments with affordable digital cameras and video processing techniques. Thus, there is noneed to factor in additional costs, and it is only necessary to add software to process the output of the surveillance and monitoring system in real-time. Moreover, a video-based method can detect fires earlier and achieve better reliability. In recent decades, video processing techniques for fire detection have been studied by many researchers [1-7]. Primarily, only color clues of the fire were...