PROJECT TITLE :

A Probabilistic Approach for Vision-Based  Fire Detection in Videos

ABSTRACT:

In the field of computer vision, automated fire detection is a hot topic. We propose and evaluate a new method for detecting fire in videos in this paper.

Fire detection algorithms based on computer vision are typically used in closed-circuit television surveillance scenarios with a controlled background. The proposed method, on the other hand, can be used not only for surveillance but also for automatic video classification for retrieving fire disasters from newscast content databases.

In the latter case, depending on the video instance, there are significant differences in fire and background characteristics. The proposed method examines the changes in specific low-level features describing potential fire regions from frame to frame.

Each of these features' behavioural change is assessed, and the results are then combined using the Bayes classifier for reliable fire detection. Furthermore, a priori knowledge of fire events captured in videos is used to improve classification results significantly.

Within estimated fire regions, these characteristics include colour, area size, surface coarseness, boundary roughness, and skewness. These characteristics are powerful discriminants due to the flickering and random characteristics of fire.

The fire region is usually in the centre of the frames in edited newscast videos. This fact is used to model the likelihood of a fire occurrence as a function of position. Experiments demonstrated the method's applicability.