A Research Of Fire Detection And Identification Based On Video Image

Fire is an uncontrolled burning phenomenon, it causes particularly great harm. In thecourse of fire prevention, fire automatic detection and recognition is very important. Thesensing device based traditional fire detection technology has many deficiencies. Not onlyvulnerable to interference from environment, but also lack of recording of situation when thefire broke out, brought difficulties to subsequent investigation. In this case, the fire detectionbased on video image come into the sight of domestic and foreign scholars. Video imagerecognition based fire detection can not only increase the speed, but for a wide range of firemonitoring also has a great advantage, you can also record the situation when the fire brokeout.Visual Attention is popular in the field of robotics and image processing techniques. Ithas been rapid development in recent years. By simulating the human visual system, visualattention mechanism can help us quickly find the "notable" part from a lot of complex imageinformation, and find critical information to the image, removing a lot of useless informationat same time. It can significantly speed up the image processing speed and accuracy of imagerecognition.This paper combines the video image based fire detection and recognition with visualattention mechanisms. The method of using visual attention mechanism to preprocess thevideo image, not only makes up the disadvantage of the traditional fire detection method, butalso significantly improves the speed and accuracy of fire detection and recognition.In order to combine visual attention mechanism with the video-based fire detection andrecognition, this paper were studied the following aspects:1. The use of visual attention mechanism in fire image preprocessing. This paper willcombine visual attention mechanism with the fire detection. First use attentionmechanism in the fire image preprocessing and make improvements withanalyzing the fire detection and recognition. The impact of time factor onsignificant is also discussed. This paper also discusses how to generate the saliencymap in the same background or the similar background. 2. Fire suspected area extraction. In analyzing the characteristics of the method ofthis paper, and the objective requirements in visual attention mechanisms, thispaper choose the single image based fire suspected region extraction method. afteranalysis the existing single image based fire suspected region extraction method,this paper finally adopted flames extraction based on RGB space, flames extractionbased on HSI space and smoke extraction based on RGB space, two suspected fireregion extraction meet the shortfall with each other, smoke suspected areaextraction and then make supplementary.3. The fire feature selection. This paper analyzes the physical and mathematicalcharacteristics of flame and smoke, taking the way the fire suspected regionextraction into account as well, and finally chooses the histogram of color, edgecolor moments and texture features as a basis for judging the fire.4. Using support vector machine as a fire detection approach. Based on the highdimension and small sample of the experimental data, Using support vectormachine SVM as fire identification method. Design experiments for various typesof kernel functions, and ultimately selected a mixed linear function kernel andradial basis function kernel method.The video image based fire detection and recognition, this paper studied, using a visualattention mechanism to improve the recognition accuracy and speed. Using a mixed linearfunction kernel and radial basis function kernel method significantly reduces the falsenegative rate of fire recognition.