Information Technology Reference
In-Depth Information
Fire Detection in Color Images Using Markov
Random Fields
David Van Hamme 1 , 2 , Peter Veelaert 2 , Wilfried Philips 1 , and Kristof Teelen 1 , 2
1 Ghent University/IBBT (IPI)
2 University College Ghent (Vision Systems)
Abstract. Automatic video-based fire detection can greatly reduce fire
alert delay in large industrial and commercial sites, at a minimal cost, by
using the existing CCTV camera network. Most traditional computer vi-
sion methods for fire detection model the temporal dynamics of the flames,
in conjunction with simple color filtering. An important drawback of these
methods is that their performance degrades at lower framerates, and they
cannot be applied to still images, limiting their applicability. Also, real-
time operation often requires significant computational resources, which
may be unfeasible for large camera networks. This paper presents a novel
method for fire detection in static images, based on a Markov Random
Field but with a novel potential function. The method detects 99.6% of
fires in a large collection of test images, while generating less false posi-
tives then a state-of-the-art reference method. Additionally, parameters
are easily trained on a 12-image training set with minimal user input.
1
Introduction
Fire detection is an important component of industrial and commercial site
surveillance systems with regard to personnel and material safety. Nearly all
of the currently employed systems rely on dedicated sensors and manually acti-
vated fire alarms. To detect fire as early as possible, a combination of different
sensor types is often made, linked by sensor fusion methods to improve reliabil-
ity. Examples of such techniques include Bao et al. [1], who use temperature and
photo-electric smoke sensors, and Li et al. [2], whose techniques rely on multi-
spectral cameras. These systems however, are impractical or too expensive for
covering large sites, especially outdoors, due to the required sensor density. A
cheap and effective alternative is the use of computer vision-based techniques
in conjunction with digital cameras or CCTV networks. The main advantages
are the large coverage area offered by a single camera, and the possibility of
integration with existing surveillance camera systems.
The state-of-the art fire detection methods in computer vision typically consist
of two main parts, modelling the most characteristic aspects of fire in video.
The first aspect is spectral information. All methods employ a color filter of
some sort, usually based on a fixed set of rules. The second aspect concerns
the temporal dynamics of flames, often combined with spatial characteristics.
The spatio-temporal modelling of fire in video was first described by Healey
 
Search WWH ::




Custom Search