A real-time method for smoke detection in monitored forest areas

An improved real-time method for smoke detection based on motion and color analysis is presented. Smoke detection in video surveillance is very important for early fire detection. In general, it is assumed that smoke is a low frequency signal, which may smooth the background. However, these conditions can be met by some pure-color objects, and smoke also produces high frequency signal on the edges. An improved smoke detection method was proposed in order to prevent these problems. The presented technique uses well known background subtraction for identifying of moving regions as well as novel color based algorithm for smoke detection. Detection efficiency is improving using a classifier that provides the identification of smoke-like regions. The main advantage is that it is not required to process each frame of the video signal for sufficient accuracy. Sampling rate was set to a few second in order to find the optimum value between performance and system requirements. Performance of the proposed method was evaluated using different kinds of the input signal, where satisfactory results were achieved