Flame and smoke detection method for early real-time detection of a tunnel fire
Introduction
At the present time, increasing urban areas by developing underground space is an important subject. Based on these demands, tunnel development is proceeding to reduce transportation time and to achieve efficient use of space. Maintenance and management of these tunnels are now important, due to increased demand for greater use and the needs for construction of tunnels. Therefore, because of increased usability of tunnels, the prevention of fire and emergency evacuation becomes important matters to consider.
In particular, when fire occurs in a tunnel, there is a serious possibility of massive injuries to those trapped inside. In addition, substantial property damage is always expected. There were some cases of accidents recently: a missile propellant explosion at Dalseong tunnel in Daegu, Korea (2006); the Mont Blanc tunnel fire (1999) in France-Italy; and the Tauern tunnel fire (1999) in Austria, in which there were large numbers of casualties and huge property damage [1].
The cases above illustrate the reasons why an early-stage detection system is required, to reduce such losses. Therefore, we try to minimize such damage by developing a device that detects fire as early as possible.
In the case of forest fire detection, it is possible to detect the fire by distinguishing between the color of the forest (green) and the fire (red) [2], or by using the difference between sequential images to detect the rapid formation of smoke [3]. A detection system for forest fire generally uses a stationary image. Tiny moving elements such as birds can possibly be ignored. In addition, when using color information, it is easy to detect both green forest and red fire, and even more easily at night.
However, when applying these types of algorithms in tunnels, detection of fire is very difficult, due to elements such as moving cars, tunnel lights, and other various situations. Therefore, it is necessary to adopt different approaches to forest fire detection and tunnel fire detection, using different algorithms. As regards fire detection in a tunnel, a study of the differences between a normal situation and a fire situation involves comparison by histogram [4], or detection of cars and/or trains with irregular overheating [5], or monitoring of irregular situations in the tunnel [6]. However, these studies do not have solutions for fast vehicle movement and false detection caused by lights of vehicles.
There are also studies to detect fire in tunnels using models including neural-network [7], a statistical color-model [8], and studies to detect smoke by a color-model and smoke expansion character [9]. However, the authors of these articles had problems related to the training period, real-time processing burden of complex calculation, and the possibility of color-shifts by lights in tunnels. Therefore, none of the algorithms previously presented are robust and flexible enough to handle all problems typical of automatic video fire detection.
The weaknesses described reduce the possibility of fire detection and must be rectified before these methods can be applied commercially.
Marbach [10] summarizes these problems as follows:
- •
Lighting conditions (day and night, artificial lights, light reflections, shadows, a car's front and rear lights).
- •
Image quality (poor camera resolution, poor camera contrast, poor signal transmission, dirty lens, vandalism affecting the image quality).
- •
Scene complexity (moving objects and people: different velocities and sizes, automobile's exhaust gas).
- •
Processor performance (real-time detection, processor speed and memory).
- •
System installation (friendly configuration and parameterization).
Because of these problems, some fires are not detected and consequently the systems may not be considered reliable. Thus, great flexibility and high reliability are required for flame and smoke detection algorithms, to reduce the false alarm rate and to decrease the alarm reaction time. Moreover, detection algorithms must neither perturb performance nor reduce the quality of the monitoring and storage task.
This paper describes the use of images obtained in a tunnel to produce a rapid and early fire detection algorithm, and proposes two algorithms that complement each other to solve the problems mentioned above. In Section 2, we present a brief review of the proposed algorithm. In Section 3, experimental results based on Section 2 are presented. In Section 4, the experimental results are discussed, including results of flame and smoke detection, and real-time issues.
Section snippets
Flame and smoke detection algorithm
Inside a tunnel, there are light and smoke factors such as car and tunnel lights and automobile exhaust fumes. As these factors may be the cause of faulty fire detection, it is necessary to have a more robust, accurate, and faster algorithm to distinguish them. Therefore, in this paper, we propose a flame and smoke detection algorithm that minimizes false detection and operates in real-time.
The proposed algorithm consists of two internal algorithms, namely; Flame Detection Algorithm (FDA) and
Experimental results
Considering the execution of the proposed algorithm in real-time, we conducted tests using various types of video clips. We used real-world CCTV videos operating in a real-world tunnel (excerpts from Dalseong tunnel fire in Daegu, Korea) and from a real-world fire test conducted by Fire Defense Department (FDD) at Korea Institute of Construction Technology. Table 1 shows basic information about the videos used in the tests. All videos have a spatial resolution of 320×240. We can obtain stable
Conclusion
Human observation of CCTV in a tunnel for 24 h is a very difficult task. Thus, when a flame and smoke detection and warning system using applicable image processing is used, fire detection is more convenient and it is possible to minimize damage when human observation is not available.
Each of the two algorithms is applicable in a different situation. The flame detection algorithm detects flame by comparing an image of the normal state and an input image using color information, while the smoke
Acknowledgments
This work was partly supported by Safety Management NETWORK of Infrastructure funded by The Korea Infrastructure Safety & Technology and partly supported by the Seoul Research & Business Development Program (CR070048).
References (12)
- et al.
Fire detection using statistical color model in video sequences
Journal of Visual Communication and Image Representation
(2007) - et al.
An image processing technique for fire detection in video images
Fire Safety Journal
(2006) - Alessandro G. Colombo, Nedies project: lessons learnt from tunnel accidents, in: JRC—ISIS EUR Report on March...
- T.H. Chen, C.L. Kao, S.M. Chang, An intelligent real-time fire-detection method based on video processing, in: Security...
- V. Cappellini, L. Mattii, A. Mecocci, An intelligent system for automatic fire detection in forests, in: Image...
- S. Noda, K. Ueda, Fire detection in tunnels using an image processing method, in: Vehicle Navigation and Information...
Cited by (80)
A study of engine room smoke detection based on proactive machine vision model for intelligent ship
2024, Expert Systems with ApplicationsAdaptive modified ant colony optimization algorithm for global temperature perception of the underground tunnel fire
2022, Case Studies in Thermal EngineeringCitation Excerpt :However, since some serious tunnel fire disasters have occurred in the past years, tunnel fire safety has been paid more attention [6–10]. The underground tunnels are usually used as road and subway lines as well as the gathering place for many types of cables, thermal and gas pipelines, and so on [11–15]. Consequently, the size of the underground tunnel is usually large, and once fire occurs in the underground tunnel, it is very important to predict its global temperature, which can be used to estimate the dangerous region for making an effective firefighting plan.
Flame detection by heat from the infrared spectrum: Optimization and sensitivity analysis
2022, Fire Safety JournalUFS-Net: A unified flame and smoke detection method for early detection of fire in video surveillance applications using CNNs
2022, Journal of Computational ScienceCitation Excerpt :In [66], a flame and smoke detection method was proposed using flame and smoke characteristics such as color, chaotic (irregular) motion, spatiotemporal wavelet analysis, contrast analysis. In [67], a method was proposed for flame and smoke detection in tunnel environments using color and motion information. In [68], a combination of two methods was proposed for nighttime flame detection and daytime smoke detection.
An intelligent tunnel firefighting system and small-scale demonstration
2022, Tunnelling and Underground Space TechnologyDetection of vapors from overheated PVC cables with modified sea urchin-like ZnO for fire warning
2022, Sensors and Actuators B: ChemicalCitation Excerpt :To prevent electrical fires as early as possible and greatly reduce property damages and casualties, investigating and developing advanced detection technologies is greatly significant and highly urgent [3]. Various smoke, temperature, flame, and gas detectors for sensing different fire characteristics have been studied and developed [4–7]. Of these detectors, gas sensors hold great potential to achieve earlier detection of fires, given that various toxic and hazardous gases and vapors have been generated and released before the other fire characteristics including smoke emerge due to pyrolysis or smoldering of combustible materials [8,9].