Video Based Fire Detection with Color and Geometric Features of Flames

Recently, video based flame detection technologies have become an important approach to the early detection of potential fire disasters. The recognition accuracy of most of the existing approaches for video based flame detection is not satisfactory. In this paper, we develop a new video based flame detection algorithm that can accurately recognize flames in video sequences based on color and geometric features extracted from regions that may contain flames in the video images. In addition to the traditional color and morphological features of flame regions, we develop a new feature to describe the color features of flame regions. All these features are then processed by a support vector machine (SVM) based classifier to determine whether a video sequence contains flames or not. Our testing results show that this new approach can effectively detect interference sources that are similar to flame regions in color and shape and achieve a high detection rate and good reliability.