Study Of Early Fire Recognition Method Based On Multi Mode Image

In recent years, with the development of digital image processing technology,image-type fire detectors are extensively studied by many scholars. Because such kind ofdetector has the advantage of applying to the indoor and outdoor environment with largespace and complex structures, and it either can not be affected by dust and moisture. Onthis basis, we propose an early fire recognition method based on a variety of image patternfeatures fusion. This early fire identification and warning worked with the foundation ofVS2010platform, using OpenCV and video capture card SDK development library toprogram.Firstly, by the analysis of various image pattern features and detection rangeconfirming the different combination of image modes for day and night to better graspearly fire flame image and give corresponding hardware information. Secondly, accordingto the characteristics of early fire flames said that sharp corners and other features can beregarded as a feature criterion of early fire flame. Thirdly, use different digital imageprocessing techniques (such as filter, threshold, and corrosion expansion etc.) preprocessdifferent image modes, then get different modes binary images separated from flame andbackground and corresponding features criterion values. Then, fusing multi-mode firefeatures by BP neural network, using the feature criterion value obtained from a largenumber of fires and interference experiments as learning samples to train well-builtnetwork, and then get the right value of input and hidden layers. Next, create a dialog boxin VS2010and give the design process of main program and subprogram. Integrating andcomputing multi-mode image of early fire feature criterion and the value of BP neuralnetwork, then comes the probability of fires. Finally, under the conditions of day and night,using three kinds of burning material and corresponding light interference experiments atdifferent distances get to the conclusion that image-type fire identification study also needsfurther improvement in the future.