煤层气开采远程视频监控系统设计

发布时间：2018-03-23 15:00

本文选题：煤层气开采　切入点：前景检测　出处：《大连理工大学》2015年硕士论文

【摘要】：我国煤层气开采产业发展迅猛,因此对现场安全情况的监控也就显得尤为重要。为解决人工巡检煤层气开采现场耗时费力等问题,设计了基于改进核密度估计(KDE)前景检测算法的无线远程视频监控系统。首先,对当前的运动目标检测算法做了综述,分析比较了帧差法、高斯背景建模(GM)、混合高斯背景建模(GMM)和核密度估计等算法,发现KDE算法更为适合本项目面临的抽水机往复运动等动态背景环境。接着针对KDE算法实时性差的问题,提出了改进KDE算法。本文算法将背景差分(BS)与核密度算法结合,先用背景差分融合三帧差算法过滤掉非动态背景区域,对于动态背景区域应用核密度算法检测前景。为了兼顾准确性,在前景分割时提出了一种新的动态阂值求取方法,第二背景更新采用随机邻域点更新策略,仿真结果显示：本文算法与核密度估计法和背景差分核密度(BS-KDE)算法相比,准确性更高,平均每帧算法耗时分别降低了94.1%和15.38%。其次,组网方式上,在3G网络上建立虚拟专用网络(VPN)通道来构建网络。在采集端的设计上以FPGA和ARM处理器为基础,工作重点在ARM上程序设计,基于C语言采用多线程方式协调图像采集、处理等工作。视频监控软件则采用C++语言设计,包括预览模块、回放模块、日志模块及参数设置模块。能够及时接收现场异常情况图像并报警,做到了人机交互较为友好。最后,测试了网络性能和算法在ARM上应用效果。网络延迟在100ms以上,为保证数据准确到达,采用包头附加数据的方式定义图像帧,其中包头包含数据长度和站点号等信息。传输图像帧时,利用TCP流协议,图像帧被自动分包发送。在监控端利用多线程,按照包头的数据长度接收各个站点图像数据。在室内外环境中对准确性和实时性做了测试,结果显示,本算法能够准确检测到大小运动目标,但由于ARM内存等资源限制,算法耗时是仿真结果的3到9倍,所以本文算法实时性还需进一步提高。
[Abstract]:China's CBM industry is developing rapidly, so the monitoring of the site safety situation is particularly important. In order to solve the problems of mining coal seam gas field manual inspection is time-consuming, the design is improved based on kernel density estimation (KDE) wireless remote video monitoring system for foreground detection algorithm. Firstly, moving target detection algorithm for current do the review, comparison and analysis of the frame difference method, background modeling Gauss (GM), mixed Gauss background modeling (GMM) and kernel density estimation algorithm, KDE algorithm is more dynamic environment background of pump reciprocating motion for this project. Then, facing the problem of KDE algorithm is proposed. The improved KDE algorithm. The algorithm of background difference (BS) combined with kernel density algorithm, using background difference difference algorithm to filter the non dynamic background region merging three frames for the dynamic background region using kernel density algorithm. Measuring prospects. In order to take into account the accuracy, obtaining a new dynamic threshold value method is proposed in the foreground segmentation, background update second randomized neighborhood point update strategy, the simulation results show that the proposed algorithm and kernel density estimation method and background difference kernel density (BS-KDE) algorithm, higher accuracy, the average the algorithm of frame time was decreased by 94.1% and 15.38%. second, networking, virtual private network 3G network (VPN) channel to construct a network. In the design of acquisition terminal based on FPGA and ARM processor based, focus on the ARM program design, C language uses multiple threads to coordinate based on image acquisition. Processing. Video monitoring software is designed by C++ language, including the preview module, playback module, log module and parameter setting module. To receive timely field anomaly image and alarm, achieve human-computer interaction is Friendly. Finally, the application effect of the network performance and the algorithm on the ARM test. The network delay in more than 100ms, in order to ensure accurate data arrival, the image frame definition of Baotou additional data, which contains the Baotou data length and site number and other information. The image transmission frame, using TCP streaming protocol, the image frames are automatically sub sent. In the monitoring terminal by using multi thread, according to the Baotou data length received image data. Each site in the indoor environment to do the test, the accuracy and real-time results show that this algorithm can accurately detect the size of moving objects, but because the ARM memory resource constraints, algorithm of time is 3 to 9 as the simulation results, so the real-time performance of the algorithm is further improved.

【学位授予单位】：大连理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TD82;TP277

【共引文献】