| As the number of image and video fire detection methods has increased over the past decade,video fire detection and analysis techniques have matured.Compared with the traditional transmission detection sensor,which has the disadvantage of transmission delay,the image processing based on image sensor has a fast response speed,which can greatly shorten the detection time.However,most single-sensor fire detection systems are difficult to eliminate complex disturbances such as reflections,color and jitter-like objects,and noise when detecting fires in real-world environments.In order to overcome these shortcomings,this paper uses a multi-sensor video flame detection technology combined with infrared sensors.Ordinary vision sensors involve visual characteristics,and infrared sensors involve thermal radiation characteristics to more accurately detect fires.The main research contents of this paper are summarized as follows:1.Analyze the type of cabin flame in ship fire,the dynamic and temperature characteristics of the flame,and the kinematics and empirical model of the flame.First,refine the type of cabin fire,and give the location of the fire,the cause of the fire and the consequences.Secondly,based on the above,the analysis of fire characteristics is mainly based on its dynamic characteristics and temperature characteristics.Then,kinematics equations and empirical models are given based on the analysis of the characteristics.Finally,based on the diversity analysis of fire characteristics,the conclusion is drawn about the importance and necessity of multi-sensor fire detection.2.Analyze and summarize the commonly used video flame detection technology.Firstly,the color detection model is analyzed.Combined with the characteristics of each color space,a multi-scale color model is given,which has a good extraction effect on fires in different scenes.Secondly,the classification of three kinds of moving object detection technologies commonly used at present is discussed,and the applicable scenes and detection effects are compared.The Gaussian mixture model extracts the contours of moving objects more completely and clearly.Finally,the flame characteristics under Fourier stroboscopic and discrete wavelet transform are analyzed.It is found that the flame and energy domain have special texture and flicker characteristics,which can be used as an important feature of flame.3.For a single sensor,the flame detection capability is insufficient,and a multi-sensor flame detection method is proposed.First,in the visible range,the spatial differences,center estimates,dynamic features,and texture features of the flame are analyzed separately.The feature is fused by an extreme learning machine(ELM)and combined with a classifier to determine the probability of detection of the flame.Secondly,in the infrared light range,the three characteristics of the gray space difference(normalization),the difference coefficient of variation and the confusion of the bounding box are analyzed respectively.The three flame characteristics are used to propose the local flame risk probability model to achieve the flame.Detection.Finally,image mapping is performed by image registration method to match visible light and infrared video to a uniform coordinate area.On this basis,the global flame classifier model is obtained by combining the visible flame detection probability and the infrared flame probability model.4.Finally,according to the analysis of the above chapters,describe and structure the flame detection system,and implement this system.The system mainly introduces the three detection interfaces and sub-interfaces used.Using this system to simultaneously process visible and infrared flame video frames and fuse detection,the detection method has higher accuracy and better reliability. |
