earticle

영어

The scarcity of automatic fire detection systems continues to be a problem that needs a serious attention in order to save human lives by preventing injuries and/or deaths. The newest innovations are continuing to use cameras and computer algorithms to analyze the visible effects of fire and its motion in their applications. As their approaches present some drawbacks when working in spatial domain, the main difficulty is still to identify objects if they do not occur at the expected position. In this paper, we present an improved fast and robust algorithm for detecting fire/smoke in a cluttered scene from a pair of cameras. The input images are first segmented according to a pre-determined disparity threshold map and the real-time disparities of fire. Binary image processing techniques are used to reject noise introduced into the segmented images through low-resolution disparity calculations which consequently can lead to the gain of clearer results. In order to reduce the false alarms, a new segmentation method used in this approach shows that segmented images using stereo-vision are more accurate than those segmented using color approach for the overall detection. The segmented images are then used for image feature extraction for a Fuzzy-neural network classifier to help the system to generate a warning in case fire/smoke is detected.