Analysis And Design Of Intelligent Monitoring System Based On Computer Networks

As an effective device for protecting people's lives and properties, monitoring system has become a new research and development hotspot. Such questions as how to use present resources and technology to realize the tracking, localization and recognition of mobile objects, how to detect motion and make alarm by image analysis, and how to communicate with each other over networks among the parts of monitoring system will be the research directions of digital video monitoring systems.In this paper I analyze and design an intelligent image monitoring system based on networks. The monitoring system is divided into two subsystems: monitoring station subsystem and control center subsystem. Monitoring station subsystems capture, store and process images, while control center subsystem is in charge of transaction management and image recognition. Subsystems communicate with each other over networks. Knowledge about digital image processing and pattern recognition is applied to do image preprocessing and face recognition about video image sequences. In monitoring stations, images can be filtered, smoothed and noise reduced via differentiation image and digital morphology methods. By these methods, mobile objects can be extracted from monitoring images and detected the location and size. In control center subsystem, further image localization and recognition of image will be done. The face in image can be localized by feature value method. To get effective recognition, I use geometric feature method, such as eigenface matching and the location, shape and grayscale distribution of eyes in face, to detect face and eyes. I also use singular value decomposition method to reduce dimensions of face and extract features. When in recognition, three-layer BP neural networks will be adopted. The result of SVD is the input and the number of classification equals the number of output-layer nodes. The parameters of BPNN have vital important effect on the study rate and convergence performance, therefore there are widespread discussion and research about the selection of neural network parameters. Here I give some references to the selection of the number of hidden-layer nodes and learning rate. The information transmitted among subsystems is video image and monitoring alarms. Monitoring stations send alarms to control center, and then require the image be recognized. After control center does the work, it sends response back to the monitoring station at a given time. The video image information can be classified into real-time stream media and stored image file. As to the former, stream-oriented transmission is adopted and x stream media protocols are employed. Presently there are RTP, RTCP, RTSP and RSVP supporting stream media transmission. RTP transmit data, and RTCP provide feedback of data transmission quality. RSVP can reserve network bandwidth to guarantee quality of service. RTSP remotely control data transmission. Generally control information is transmitted by HTTP and multimedia data are transmitted by RTP/UDP. As to stored image files, mostly bitmaps and short-time clips, control center accesses files by FTP. It downloads and stores files in database server as backup.Finally, I do object-oriented analysis and design of intelligent monitoring system based on networks, making system model by UML.