The Research Of Forest Fire Recognition Algorithm Based On Video Streaming

Forest is one of the important natural resources on the earth,but the fire often happens in this area.It is known that the fire happening in the forest is sudden,random and often causes great loss in a very short time.How to predict the forest fire timely and accurately has become an urgent problem to be solved.Nowadays both at home and abroad forest fire detection is just at the stage of manually monitor screen.Moreover for the high rate of false alarm,the automatic monitoring and control technology is still in a immature application.Therefore,the research of forest fire recognition algorithm based on video streaming has a greatly important and practical significance.We know that it appears a lot of smokes during the early occurrence of the fire.So this paper mainly does a research in the smoke detection of video streaming.The process mainly includes preconditioning of the input video image,the detection of moving target in video image,the extraction of texture feature in smoke image and the classification and recognition of support vector machine(SVM).Because of the complexity of the forest environment,the contents in this paper includes the steps and methods of each aspects of the recognition algorithm.The main work:(1)In the stage of preprocessing video image,the removement of the noises and the weighted average method provide a high quality image and improve the speed of processing.(2)Compared to the various common foreground extraction algorithms,we shall apply the method of moving target dection based on improved background estimation with color judgment which not only has a better ability of capturing smoke,but also can preliminarily rule out the birds,shaking trees and etc.(3)A new algorithm of image feature extraction based on the combination of multifractal and Contourlet transform is proposed.First,the image is decomposed by Contourlet transform to extract the directional subbands,and then the image texture features are described by the analysis of the multifractal characteristic parameters of each directional subband.Compared with gray level co-occurrence matrix,multifractal algorithm and others,it can be showed that the algorithm of this paper could improve the rate of forest fire smoke recognition and reduce the rate of falsealarm better.(4)According to the features of experimental data of high dimensions and small samples,the support vector machine(SVM)is used as a method to identify forest fire image.The characteristic data are normalized and the radial basis function(RBF)is selected as the kernel function of the support vector machine model.Experimental results show that the forest fire recognition algorithm based on video streaming is capable of timely and effective identification of the smoke and has a good anti-disturbing ability.