In this article, a video base early fire alarm system is developed by monitoring the smoke in the scene. There are two major contributions in this work. First, to find the best texture feature for smoke detection, a general framework, named Histograms of Equivalent Patterns (HEP), is adopted to achieve an extensive evaluation of various kinds of texture features. Second, the \emph{Block based Inter-Frame Difference} (BIFD) and a improved version of LBP-TOP are proposed and ensembled to describe the space-time characteristics of the smoke. In order to reduce the false alarms, the Smoke History Image (SHI) is utilized to register the recent classification results of candidate smoke blocks. Experimental results using SVM show that the proposed method can achieve better accuracy and less false alarm compared with the state-of-the-art technologies.

Early Fire Detection Using HEP and Space-time Analysis