Research On Underground Coal Mine Fire Detection Algorithm Based On Video Surveillance

The coal industry is a pillar industry of the national economy, due to the particularity of coal mining, coupled with most of the coal mine underground geological conditions is more complex, leading to the frequent coal mine accidents occurred, especially the coal mine fire caused by gas is the major cause of coal mine accidents. In order to detect the fire,scholars have given various fire monitoring algorithms,but for the special condition of mine fire,these algorithms are not very mature, easy to cause the misstatement and omission. In view of this situation,this paper has carried on the research to the fire monitoring underground, and the main work is as follows:1.The enhancement of coal mine dust image. In view of the videos obtained under the mine accompanied by lots of noise, low resolution and the blurred image caused by the coal mine dust,low intensity of illumination and the water pollution,the image enhancement algorithm under coal mine dust based on the theory of dark gray prior and adaptive bilateral filtering is proposed.By enhancing the video image contrast, enriching the image details, to achieve the goal of removing noise.2.The segmentation of flame image.A real-time flame segmentation algorithm based on the improved VIBE and color recognition rules is proposed.By combining the improved three frame difference method and vibe algorithm,to eliminate the hole,ghost and other problems existing in VIBE algorithm.Using HSV color model and YCbCr color model to get the flame color recognition rules, through the separation of the flame brightness information and color information, to further eliminate the fire area.3.The fire detection algorithm in coal mine. In view of the current fire recognition algorithms with problems of high computational complexity,low detection rate and low false alarm rate,combining with the special underground environment,this paper selected the flame area feature, edge feature, texture feature, flash frequency feature and roundness feature as the training samples totrain the support vector machine. Finally, use the support vector machine(SVM)based on the extracted feature to classify and recognize the flame image.Experiments show that the proposed fire detection algorithm based on dynamic target detection and SVM, has high detection rate, and can satisfy the requirements of the fire control system, is of great value to use...