Research And Design Of Fire Detection Algorithm Based On Video Image

The fire is one of the common disasters which seriously threatens people's life and property.The problem of detecting fire in time has been the important research in the field of fire prevention.Traditional methods such as temperature,smoke,light detectors have limited range of detection,and are vulnerable to outside interference,and can hardly meet the needs of fire detection in complex and specific circumstances.In order to solve these problems,based on video image processing and the theory of the shallow machine learning and deep learning,two fire detection methods based on video image are proposed.Based on the shallow machine learning,a fire detection method with multi-feature based on BP Neural Network is designed.First,dynamic detection and color model are used to separate the suspected area of fire;Then,the static characteristics such as color,shape,texture and the dynamic characteristics such as area variation and blinking are selected to compose characteristic vector;Finally,using characteristic vector as the inputs of BP Neural Network and to judge the generation of fire.During network training,the problem of over-fitting is eliminated by adding regularization factor.The experimental results show that the method based on BP Neural Network has good recognition effect and can realize the early warning of fire.Based on the deep learning,a fire detection method with sliding windows based on Residual Neural Network is designed.Using Convolutional Neural Network to extract the flame and smoke characteristics in the image automatically to identify the fire.Using residual block can avoid the problem of gradient disappearing in deep Convolutional Neural Network.Also,the sliding windows are used to locate the fire.During network training,the network parameters are initialized by transfer learning to accelerate the training process in case of fewer samples.The experimental results show that the method based on Residual Neural Network has better recognition effect.Compared with BP Neural Network,Residual Neural Network has higher detection accuracy and stronger applicability.