Study On Video Monitoring And Image Processing Of Coal Mine Dynamic Targets

The coal mine video operations have many problems such as complex environment, large noise, uneven illumination, shades, and low-level intelligence caused by much manual intervention and the high rate of false detection. Therefore, it is necessary to study technologies as image enhancement, quick query and match the extracted feature points, real-time dynamic target detection, remnants detection and target tracking, and to analyze and understand irregularities, suspicious targets and potential risks appearing in images so as to trigger the intelligent joint alarm rapidly and reasonable, providing first-hand information for the later analysis of the accident. Based on the study on the relevant technology of image processing in coal mine, the paper aims at safety problems of coal mine production, and studies on related algorithms of the video monitoring and image processing of coal mine dynamic target.In order to solve the problem of unclear picture from strong noise, reasonable extraction Local fuzzy fractal dimension(LFFD) similarity enhancement algorithm is proposed as a criterion based on fuzzy entropy. The algorithm determines a reasonable LFFD fusion based on fuzzy entropy discrimination similarity measure to adjust image contrast enhancement, and considers the application of multi-parameter in the process of measuring the similarity theory. The anchor net image processing results of Xiangshan coal mine in Shaanxi Hancheng city show that the algorithm can better reduce the noise, and enhance image contrast. As for obtain the right key component of mine equipment, support and other features and details of the texture image, it is considered that the mine is affected by noise and other environmental factors on the shooting feature edges. Thus an improved canny edge detection algorithm based on wavelet decomposition is proposed, and then a canny edge detection algorithm based on wavelet decomposition is proposed. The algorithm utilizes the wavelet transform to extract high and low frequency components of grayscale images, in order to extract more edge information of perfect feature profile, which plays an important role in accurate collection of feature point cloud.For problems such as the deterioration in the quality of surveillance images because of the coal mine underground video image interfered by dust and light, the overmuch acquisition points of video monitoring system of coal mine, and large amount of history retained data is not convenient to the subsequent search of feature image. This paper presents a Euclidean distance registration algorithm based on a correlation method. Through using information of different feature points, this method uses gradient correlation for gray information and uses descriptor correlation for descriptor information of SIFT algorithm based on the Harris algorithm, and combines with the Euclidean distance feature points to achieve exact matching. The image matching results of coal mine show that the algorithm greatly reduces the number of wrong matching points.Confront with coal mine underground video uneven illumination, large noise environmental, the objects are easily lost. The safe production of coal zone requires accurate investigation of the foreground the image. This dissertation studies moving object detection algorithm based on Codebook model(CBM).In the case of lack of information when the target motion, CBM may generates localized leak detection or measurement error and other problems. Through integrity of joint target spaces, this dissertation proposes a whole background updating algorithm on the target space based CBM. This method makes use of the analysis of the change of spatial moving targets information, looking for potential foreground in the background, and co-pixel time-domain statistical background model updating. The experimental results show that the algorithm can quickly adapt to the change of background. When processing the slowly moving targets and the targets only including the local motion information the method can reduce false positives due to the movement and ensure the integrity of target detection at the same time. Aiming at the problem of object detection by shadow interference a codeword component average algorithm is proposed based on HSV space, The algorithm constructs the codewords weighted average background model, and the RGB color space is converted into HSV space so as to updated background and remove the shadow effect. The experimental results show that, the algorithm has strong robustness for shadow removal.Aiming at problem that coal belt conveyor may damage tape and drum and other equipment. The study find the algorithm based on multiple background models, by controlling the updating speed of different models, and comparing the difference between models to judge the remnants. The detecting speed of this algorithm is slow, and the detection of "ghost" occurs. The remnants detecting algorithm based on the historical pixel stability is presented. It records several frames of pixel information before pixels which do not belong to the background codebook model so as to form historical pixel set. Counting the matching degree of current pixel and historical pixel judges the stability of pixel and then decides whether the remnants exist or not. Finally, the method is validated by the video of belt conveyor used in coal mine.For complex motion of mines dynamic targets, illumination changes, occlusion and other factors on target tracking performance, the accuracy of existing tracking algorithm based on multi-feature fusion is not high in a complex environment and most methods use a single problem determination for multi-feature fusion. The study proposes a target tracking fusion algorithm based multi-feature criterion. The method first introduces local background information to enhance the description to targets, and secondly in the multi-feature fusion process it uses a variety of criteria adaptive for computing feature weights. In the Mean Shift framework, the object tracking is performed with kalman filter. Video experimental results of coal mine in Shaanxi ZhangJiamao show that the algorithm is better than the single determination fused method with better stability and robustness, and can effectively improve tracking accuracy in complex environments.