Research And Application On The Algorithm Of Fast Fire Detection

The use of fire indicates that human beings enter the civilized world,and it is also a serious threat to human life and property security while promoting the progress of human civilization.Therefore,how to find the fire early is of far-reaching significance to eliminate the fire in the bud.Compared with the traditional temperature sensor or smoke,flame detection has wide detection range,fast response time,real-time detection of fire etc..Some algorithms of fire detection are discussed here,such as the complete flame color region extraction,the Bidimensional Empirical Mode Decomposition(BEMD)in this thesis.Aiming at the existing problems,a novel algorithm for fast video fire detection is proposed based on support vector machine,combined with the clustering and the flicker feature.(1)For the problems of the poor extraction of effective texture information and it results in lower classification results.In order to improve the accuracy of fire detection,a fire detection algorithm based on Bidimensional Empirical Mode Decomposition(BEMD)and Support Vector Machine(SVM)is proposed in this thesis.Firstly,candidate fire regions were detected based on the accumulative difference method for detecting moving targets and Ohta color space with color model of flame.Secondly,a new method combining the bidimensional empirical mode decomposition(BEMD)with local binary pattern(LBP)is proposed for texture image classification.The LBP is used to extract the features of a series of various intrinsic mode functions(IMFS)images and residual images,which are decomposed by bidimensional empirical mode from the image.Finally,the roundness,rectangle degree,height of center of gravity,texture features are input into the SVM classification.(2)For the problems of the imperfections the segmentation of the flame area and it results in missed detection.A fire detection algorithm based on fuzzy C mean(FCM)combined with the dynamic characteristics of the flame which is flicker feature is proposed in this thesis.Firstly,the moving target is extracted by using the cumulative difference.And then the candidate fire regions were detected based on fuzzy C means color image segmentation method of hill-climbing method(HRFCM)combined with the flame color features.It not only can effectively shorten the time,but also overcomes the problems that the number of traditional clustering algorithm,the initial cluster centers and cluster random blindly choose the initial parameters.Finally,the flicker frequency feature is input into the SVM classification.At the end,the experimental results show that this algorithm has better robustness and detection efficiency with response to some other algorithms of fire detection.