An Integrated Framework for 24-hours Fire Detection
In this paper, the integrated framework for 24-hours fire detection with a camera is proposed. The framework consists of four novel modules: an integration module, a flame detector with a visible-light camera, a flame detector with an infrared-ray camera,
- PDF / 1,988,447 Bytes
- 17 Pages / 439.37 x 666.142 pts Page_size
- 52 Downloads / 216 Views
Abstract. In this paper, the integrated framework for 24-hours fire detection with a camera is proposed. The framework consists of four novel modules: an integration module, a flame detector with a visiblelight camera, a flame detector with an infrared-ray camera, and a smoke detector. According to the state decided by the integration module, different detectors are selected to find fires. The flame detector with a visible-light camera determines flame patches from candidates through the cascaded classifiers, based on the color, shape, and randomness of flames. The flame detector with an infrared-ray camera finds flames, using the random movement of blob candidates. The smoke detector recognizes the smoke regions by utilizing the colors and the transparent property of smoke. The three detectors and the integrated framework are tested with numerous videos, which validates the generality and the robustness of the proposed framework.
Keywords: Integrated framework Smoke detection · 24 hours
1
· Fire detection · Flame detection ·
Introduction
According to the reports from National Fire Protection Association [1], 15 % of home fire victims have been caused by the physical disability, which ranked second among the fatal factors of home fires. Most of the victims could not avoid the death because of delayed escapes from a fire. However, because conventional fire warning systems, such as a water sprinkler and a manual warning lever, are operated only by strong fires, it becomes too late for the physically disabled people to escape from the fire. Therefore, the fire warning system for all day is essential for the physically disabled people. Among the various fire warning systems, the systems based on a vision sensor have been spotlighted due to its low price and easy installation. Therefore, there has been various research for the early fire detection with cameras. The most general fire detection algorithm is a flame detection based on a visible light (VL) camera, which can be categorized into three types: pixel-level, c Springer International Publishing Switzerland 2016 G. Hua and H. J´ egou (Eds.): ECCV 2016 Workshops, Part II, LNCS 9914, pp. 463–479, 2016. DOI: 10.1007/978-3-319-48881-3 32
464
J. Choi and J.Y. Choi
Fig. 1. Various Images of Fire. The shapes of fire are various according to scenes and material. (a) shows the fire images captured by visible light cameras. (b) shows the fire images captured by infrared ray cameras.
blob-level, and patch-level algorithms. The pixel-level algorithms find flames by utilizing pixel-wise features including colors and flickers [2,3]. The pixel-level algorithms work very fast, but they show low performance because the shape of flame cannot be considered and the classifiers with the simple pixel-wise features can be easily biased by training data. The blob-level algorithms detect flames by extracting features from blob-level candidates [4,5,7]. The blob-level algorithms show better performance than the pixel-level algorithms, but their classifiers are hard to be trained due to the various shapes of flam
Data Loading...
