Research On The Algorithm For Image-type Fire Smoke Detection

With the rapid development of economy, fires occur frequently, which has created a significant threat to human lives and property safety. As conventional fire detectors will be affected by various factors such as space, height, air velocity, dust etc, as well as the rapid development of image acquisition and computer technology, fire detection based on video sequences (image type) attracts much attention as a new type of effective early fire detection technology. In this paper, a thorough research has been made on the algorithm of fire smoke detection based on video processing.Visual features are the most obvious characteristics to distinguish between certain things. Color, turbulence phenomena and diffusion movement are the most obvious visual characteristics of the fire smoke, and those will be analyzed in the following sections. Firstly, a moving object detection method will be used to extract the suspected smoke area and this can discard all the static interferences similar to smoke. In this paper, the advantages of optical flow method and the frame difference method are made full use.A improved method based on the gradient of the HS is proposed to make up for the shortcomings of the classic HS method which is not sensitive to the detection of slow moving target.And some greater improvements have been done in the traditional frame difference method based on using the continuous several frame images to roughly locate the suspected smoke region. Taking full advantage of the merits of both approaches and compensating for the lack of the two, these two improved methods are intergraded to extract the moving object better which paves the way to the following characteristics extraction. After that, color analysis on the moving target will be made and a set of rules of the color of smoke which changes generally from light gray to black will be worked out. Any object that not meets the rules will be proved not to be smoke; otherwise, a deeper analysis on the feature is necessary. Thirdly, four parameters generalize the turbulent phenomena and diffusion of smoke; they are frequency of boundary flashing, cumulative exercise, convex, and the main direction of motion. Finally, the Naive Bays classifier based on the principle of the maximum posteriori probability integrates the four characteristic parameters of a given event described in a set of features to make a fire or non fire classification judgment, which can reduce the false alarm rate.The experimental results show that the proposed algorithm is able to make accurate and effective judgment in most scenarios smoke video.