Design Of Remote Video Surveillance System In Coal-bed Methane

Coal-Bed Methane(CBM) industry in China is developing rapidly so the security monitoring in the mining sites is very important. An wireless remote video surveillance based on improved Kernel Density Estimation(KDE) has been designed in order to solve the problems such as time and energy wastes when inspecting CBM sites manually.At First, a summary of current moving object detection algorithms are taken. Frame difference, Gaussian Model, Gaussian Mixture Model(GMM) and KDE algorithms are compared and then it’s found that KDE is more suitable than others to the background environment of reciprocating pumps and other troubles faced by the project. An improved KDE algorithm is proposed to the problem of poor real-time of KDE. The method combine the background subtraction(BS) and KDE, the non-dynamic background region is filtered by the BS and three frame difference algorithm first. And then KDE is used to segment foreground for the dynamic background region. A new method of determine dynamic threshold has been proposed when segmenting foreground region in order to gain higher accuracy. The strategy of updating random neighborhood pixel is used to renew the second background model. As the simulation experiments show, comparing with the KDE and Background Subtraction Kernel Density Estimation (BS-KDE) respectively, the accuracy of the proposed algorithm is higher and the time-consuming of average per frame image is reduced by 94.18% and 15.38%.Secondly, for the formation of networks, the Virtual Private Network(VPN) tunnels are established via 3rd-Generation(3G) to constitute the network. The design of acquisition is based on FPGA and ARM processor and the work focus on programming in ARM, multi-thread based on C language is used to coordinate image acquisition, processing and other works on ARM. Video surveillance software is designed by C++ language, and it includes review module, playback module, log module and parameter settings module. It can receive anomalous live images opportunely and show the alarm information. And human-computer interaction of the software is friendly.At last, an experiment of network performance and algorithm detect results on ARM is made. The network latency is above 100 ms, in order to ensure data arrives accurately, the image date frame is constituted by the header and data where header includes date length and site number. An image frame is divided into packets automatically using Transmission Control Protocol (TCP) when sending image data, and the monitoring receive date from every site based on the data length in header using multi-thread. The test of accuracy and real-time is taken indoors and outdoors, as the results show, small and big moving objects can be detected accurately by the algorithm. However, due to resource constraints, such as ARM memory, the time-consuming is 3-9 times as the result of simulation, so real-time of the proposed algorithm should be improved in the future.