Background Model for Video-Based Smoke Detection in Outdoor Scenes
Abstract
Introduction: Early smoke detection on videos obtained from outdoor video cameras is very important because smoke usually becomes visible earlier than flame and can be considered an objective evidence of a fire. Optical surveillance systems are cheap and therefore commonly used for smoke detection both in urban areas and in terrestrial forests where special fire watchtowers are maintained. Smoke is detected when objects on the video have specific motion, color, texture and shape. The resulting quality of the smoke detection algorithm depends on the frame resolution and on the distance to the objects being shot. Purpose: Background models should be build for the close and remote scenes taking into account the atmospheric and meteorological conditions. Results: We have studied two approaches to determining the scene depth: using the dark channel and a Markov random field, in both cases taking into account the law of direct attenuation of light waves in open spaces and the influence of atmospheric light. The method based on a Markov random field provided better results as compared to the method of the dark channel, as in the latter the pixel intensity is analyzed without considering the environment. Practical relevance: The found scene depth allows us to split the images into two groups: close scenes (up to about 500 m) and remote scenes (more than 500 m), where “close” and “far” smoke can be watched, respectively. During the experiments, we analyzed 100 images with close scenes and 100 images with remote scenes. The proposed method is efficient because different sets of smoke features are used for close and remote scenes.Published
2016-08-19
How to Cite
Pyataeva, A., & Favorskaya, M. (2016). Background Model for Video-Based Smoke Detection in Outdoor Scenes. Information and Control Systems, (4), 44-50. https://doi.org/10.15217/issn1684-8853.2016.4.44
Issue
Section
System and process modeling