Research Of Video Based Flame Detection Algorithm

As one of the natural disasters,fire affects everyday life all over the world.The traditional fire detector contains temperature,smoke,light,gas sensing and composite fire detector.The detector acquisition information are often relatively simple,seriously affected by the environment.With the development of signal processing and artificial intelligence,the fire detection system based on video image is produced.The image information of the monitoring scene is obtained by using the camera equipment,and the fire is detected automatically.With the characteristics of low cost,high detection rate,short response time,and so on,the related research has been paid more and more attention.In this thesis,firstly,the development status and theoretical research results of flame detection algorithms are summarized.This thesis studies and designs the flame video foreground detection method,analyzes and extracts the multi dimension feature to represent the flame effectively,and realizes the flame detection based on support vector machine.The method of deep learning is studied,and the convolution neural network is used in flame detection.On the basis of general convolution neural network,the depth of the network is increased.The main work of this thesis is as follows:(1)A method of flame detection in complex environment is proposed.In view of the fact that the flame color model is usually a color interval,it is difficult to describe the characteristics of flame accurately.The flame color is analyzed in different spaces,and to find the spatial distribution of flame pixel values,a more accurate color probability distribution model is established.Aiming at the difficulty of extracting flame region in complex scenes,a method of foreground extraction based on search box is proposed.The method takes full use of the color and motion characteristics of the flame,and combines color and motion inter frame correlation,and then extracts the block region which contains flame.The flame characteristics based on image blocks are studied.The characteristics of flame color saliency,spatial gradient,gradient between frames,flicker feature and motion feature of flame centroid are fused for flame characterization and recognition.According to the characteristics of the fusion,SVM is used to classify and detect the flame.And the influence of each feature on the classification accuracy is analyzed.(2)Considering the fast development of deep learning in recent years and the remarkable performance of convolutional neural networks in image classification,it is applied to flame detection and compared with the traditional pattern recognition methods.A deeper convolutional neural network structure is designed,and the classification and detection of flame are realized.The performance of the convolutional neural network is analyzed by experiments.The experimental results show that the proposed algorithm can extract the target block accurately in complex background and improve the recognition efficiency.The design of multi dimension flame feature can be used to detect flame by SVM classification.Convolutional neural network can be applied to flame recognition,but it is necessary to enhance the generalization ability by providing more targeted training data sets.