Rapid Early Fire Smoke Detection System Using Slope Fitting in Video Image Histogram

Abstract

Fire is one of the most dangerous natural/manmade disasters that endangers human life and property. Although early fire smoke detection systems have become increasingly widespread, it is particularly important to study videos of these systems to better understand the effectiveness of fire smoke detection, primarily because this will help to reduce losses from fires. Some of these systems have algorithms that tend to regard motion (such as moving people, cars and other non-smoke objects) in surveillance videos as early fire smoke, and this causes them to create false positive alarms. In attempting to resolve the problem of false positives, this paper outlines a new fast detection method that can be applied to early fire smoke. The proposed algorithm relies on color and diffusion characteristics of smoke and also counts the number of pixels in each candidate smoke region. A time window of 30 consecutive frames is defined to fit the linear change rate of smoke every 10 frames. The smoke discrimination criterion is provided through the smoke slope relationship, and a fire alarm is triggered. The algorithm tests five different kinds of scenarios and compares the detected smoke start frame against the observed smoke start frame. In drawing a comparison with the two test algorithms in this paper, this paper extracts experimental results that show the proposed algorithm has better detection speed and accuracy, and also suggests that it can effectively resolve the problem of false positives caused by moving non-smoke elements such as pedestrians and cars.