Research On Algorithm Of Fire Smoke Detection Based On Video

Traditional sensor-based fire detection system will tend to fail in some large indoor spaceand ventilation places, so researching a new fire detection method based on computer visionand video image processing is of great significance. Using video monitoring the fire happensor not in the scene has some advantages, such as it is not affected by distance space andenvironmental constraints, it response fast and detection intuitive to protect the larger area.The fire occurred early will usually produce large amounts of smoke, by analyzing the videosequence to detect the scene with fire smoke is the main research work.Firstly studied three commonly used methods in moving target detection, contrast andanalyze the respective application occasions, the advantages and disadvantages, anddetermined Gaussian mixture model as the method of background modeling and theextraction of the prospects, which is suitable for smoke detection in this article. The smokealong with the other moving target was extracted out together, and then analyzed the colorinformation in the extracted area. After a large number of observations and test statisticssummed up the law of the color of smoke images, which is that the fire smoke is generallywhite, blue, gray and black, and their three sub-value of RGB is basically equal, justdistributed in the diagonal of the RGB color cube, and the brightness is also within a certainrange. And according to these, the smoke color decision conditions were designed. The coloranalysis for the motion detection area can further rule out the moving object which does nothave smoke color features. Then after the morphological processing and connectivity analysisin the motion detection and color analysis area, obtained the suspect smoke regions in thevideo.Next in order to further improve the accuracy of the smoke detection, looked for threedynamic characteristics of smoke in the suspected smoke area. The smoke in addition to thecolor is different from other moving objects, it also has its own unique dynamiccharacteristics. In this paper there are three features were extracted, which respectively were the growth of the area in the smoke spread, irregular contour feature of the smoke region andthe background to blurred when smoke appeared. And those three dynamic characteristicswere fused by a BP neural network to determine smoke or not. The neural network designedin this article is a three-network which has seven hidden layer nodes, three inputs and oneoutput. And the training and extensive testing found that it can be good to recognize smoke.Finally, in order to verify the effectiveness of the proposed algorithm, there were anumber of different scenarios videos including smoke and non-smoke and as well as smokeand disruptors coexistence had been tested. Test results show that the video smoke detectionalgorithm in this article can identify smoke in video accurately, real-time and effectively, andit also has some anti-jamming capability.