Efficient visual fire detection applied for video retrieval

In this paper we propose a new image event detection method for identifying fire in videos. Traditional image based fire detection is often applied in surveillance camera scenarios with well behaved background. In contrast, the proposed method is applied for retrieval of fire catastrophes in newscast content, such that there is great variation in fire and background characteristics, depending on the video instance. The method analyses the frame-to-frame change in given features of potential fire regions. These features are colour, area size, texture, boundary roughness and skewness of the estimated fire regions. Because of flickering and random characteristics of fire, these are powerful discriminants. The change of each of these features is evaluated, and the results are combined according to the Bayes classifier to to achieve a decision (i.e. fire happens, fire does not happen). Experiments illustrated the applicability of the method and the improved performance in comparison to other techniques.