Fire detection using color and motion features in video sequences

In this paper a real time fire detection system was designed and implemented using moving, colored images taken from natural environment. There are two main steps in the implemented fire detection system. In the first step, areas that have flame colour properties were detected and in the second step, the detected areas were assessed for the motion properties of flames. Flame samples taken from various fire images were analysed with different color spaces. The result of this analysis showed that fire colors can be automatically detected using triangle correlations between Y-Cb and Cb-Cr binaries in YCbCr color space. Areas with all three components supply the related correlations were marked as flame areas and then motion properties of connected components consist of these areas were analysed. During the motion detection, characteristic displacement and vibration features of flame motion were defined. Components having all these properties were accepted as flame components. Developed fire detection system was tested with real forest fire videos and non fire videos including objects that have the same colour properties as fire.