Full-Scale Video-Based Detection of Smoke from Forest Fires Combining ViBe and MSER Algorithms

Abstract

Smoke, as a prominent character of combustion, is widely regarded as a signal of forest fire. Existing in a video-based smoke root detection methods on rely the distance between smoke and the lens, which is one of the most challenging parts. In relatively close distances, the dynamic region extraction method not only presents simplicity but also provides clear outlines and shapes, which is good for the smoke root extraction. However, when the distance increases, the advantage of this algorithm decreases and the rate of leak detection rises. To solve this challenge, this study developed a new algorithm which adopts Bayesian theory to combine the ViBe algorithm with the MSER algorithm. The likelihood functions are replaced by a small database, storing the descriptors of each frame and being updated in real time. The experiments demonstrate that the new method produces more complete shapes of candidate smoke root regions and lowers leak detection rate in full scale, compared with the ViBe results and MSER results, respectively. These improvements suggest that it can detect smoke in the forest more accurately.