Fire Flame Detection Based On Support Vector Machine

Considering the shortage of traditional fire detectors in special facilities such as high raise building, it is now believed that the video fire detection methods based on surveillance system and machine vision will be widely used.The principle and characteristic of the fire flame detection technology were summarized and the existing fire flame detection algorithms were researched. Furthermore, a new fire flame detection algorithm based on support vector machine was proposed in this paper.It is supposed that the video frame obtained by camera were separated and to form images sequence firstly. The outer-flame-color region in the original image was segmented by using RGB model, and the suspected fire flame region was detected by segmenting the inner-flame-color region in the first segmented result using HSI model. Then the circularity, the amount of sharp angles, area proportion of red and green components, area variance rate, related coefficient and flicker frequency in the suspected region were calculated and analyzed. At the same time, the simulation experiments were performed. At last, the model parameters of support vector machine were determined by experimental method. On this foundation, the support vector machine classifier was built. The extracted flame features were assembled as the input vectors of the SVM classifier and the feature data can be classified and recognized by SVM. The experiment results show that the algorithm has comparatively high positive alarm rate. Besides, when the detection environment and burnable material changed, the separating hyper-plane can be generated automatically by training SVM and it can separate the sample data accurately. So it is self-adaptive as well...