Research And Implementation Of Analysis System For Moving Information Based On Video Stream

The direction of intelligent computer has been established in the rapid developingprocess of information. Under the increasingly serious conditions of the global fight againstterrorism, the job of security guard has become more and more difficult, then the intelligentmonitoring system can provide great assistance and even to some extent can replace peoplejob security. Intelligent monitoring system is on the installation of a "brain" in the generalcomputer monitoring system, it can automatically analyze a particular type of things afterstudying. General monitoring system's main role is to help security guarders who monitorthe scene to ensure no anomalies occurred, but the security guarders are required to watchthe whole monitor screen, intelligent monitoring system can't do it like that and cananalysis automatically monitor screen what is happening and whether there is abnormalcircumstances, System requirements in accordance with the corresponding action.This choose of the article title is based on the combination project of production, studyand research in Guangdong Province Office of Education - Content-Based Retrieval ofNetwork Video Monitoring Machine (Item Number: 2007B090400031) and thespecial-fund project by Development of Information Industry of Jilin Province -Content-Based Retrieval of Network Video Surveillance (contract number: 2007042), theabove-mentioned projects belong to the field of Intelligent Video Surveillance. Theintelligent video monitoring system based on R & D, the main analysis of the movingobjects in monitoring screen, the system can know that moving objects are which typesobject and what action to do. Test based on the video surveillance of the road, test resultsshow that the design has a certain practical value and reference. This article related to studymuch basic knowledge and read a lot of related papers at home and abroad, research anddesign the relevant algorithms to fit the requirements of the system, and to carry out theimplementation and simulation tests.This paper deals with motion detection, visual image processing, feature extractionand image pattern recognition and related technologies,, every part selection andimprovement of technical characteristics have been all kinds of comparisons which is basedon algorithm's time complexity and space complexity and considered the program to runthe strength of environmental factors such as adaptability, design a set of the analysissystem of moving information based on video streams: through the video sequence to detectthe moving object information, By use of computer vision technology for image processingto get moving targets binary-value regional map, mark the map in connectivity and extract the geometric features and the behavior features of images, eigenvector composed bysupport vector machine (SVM) classifier put into each exercise objectives to the knowncategory and analysis of exercise behavior.Motion detection work is the basic part of the system in this paper, the background ofthe monitor is established by Gaussian mixture model, then the system extract movinginformation by differencing the background model, and update the background model inreal-time detection of moving targets. This technology can effectivelydetect moving targets.but the paper also provides time series of difference and single-Gaussian Detection as acomparison, because motion detection is great sensitive to external influences, mixedGaussian has a strong ability to remove environmental noise and the system test is moredesirable to the extraction of moving regional map.Moving information which is extracted from video streams by motion detectionincludes many noise, these noise in the computer is no distinction with our concernedmoving goals, but their existence will seriously affect the results of the system's analysis,we must make use of de-noising algorithm to eliminate these noise, this paper, by use ofnoise characteristics of the smaller area, using 8-neighborhood decision-making template tofilter out these small image region, can retain a large area which is concerned to our targetsports information of the image data; these images information also can not satisfy theneeds of the paper, there is a lot of flaw in moving objects region map which includes a lotof "glitches" and "empty" elements, that required the use of morphological operators to dealwith the image information, then through the connectivity tests marked by the number ofmoving objects, and establish the corresponding data structures of the moving goals, tomaintenance the moving objectives and their behavior. In according with the map for thephysical features of each moving object are extracted; Secondly, the use of the movingtarget area map of two adjacent detections is to extract behavioral features, and composemoving feature eigenvector of the physical features obtained, and normalized to provide tothe image analyzer.Final job is based on physical features and behavioral features of the moving objects,the support vector machine (SVM) that has been studied for some time is used to analysisthe feature eigenvector for object recognition and behavior classification of moving goals.The theoretical foundation of SVM is structural risk minimization, SVM chooses a suitablekernel function to converse space dimension, put non-linear problems to transform intolinear separable problems, and find the best plane of classification. SVM solves theproblem that the training samples are less actually, a small error classifier is got by limitsamples training can, it's a smaller test error to the independent test set, SVM has been thegreat popular classification techniques in the field of Intelligent Recognition Technology.In this paper, a large number of experimental results show that the system can reach ahigher recognition rate and the stronger analytical capacity after the studying of SVMthrough the small number of samples. In this paper, the constraints operating platform can be fully taken into account in theprocess of implementation algorithm, as well as the algorithm robustness, theimplementation of core modules is to complete the work at PC by C language, through theVisual C++ system simulation interface to assist the completion of the test situation. Lookat the result of the simulation tests, the function of the system is more satisfactory, thesystem can be understood basically the behavior of people or vehicles in video monitoringscreen, whether the action is unusual and so on? However, the original intention of thisarticle are achieved to run this system in an embedded platform, if the embedded platformwas defined, we encounter a lot of issues in the process of transplant the algorithm,including the concerned question in this article whether or not to meet the computingcomplexity of embedded request, even if using C language to implement the core modules,it will also have a lot of questions, these questions can only be resolved at that time; Lookat the system functions there are a number of modules to be further research andimprovement, such as support vector machines set up the need for further research andimprovement, the experimental tests of this paper is relatively simple, which are based onan analysis of two types of moving goals, we must consider how to construct many types ofclassifier. At follow-up to give further consideration to more constraints, extract moreimage features in order to analysis more complicated behavior of the things, and improvethe intelligence system.In this paper, the result of the system simulation can show there must be a greatsuccess on a development path of the intelligent video monitoring system, Intelligentmonitoring system and the direction of the overall development of the computer to maintainconsistency, but the function of monitoring system for the monitoring system can not onlyimprove the functionality and efficiency and may even replace the security personnel tocarry out security work; In computer's intelligence of view, it has even become the drivingforce of change the way of the people life, So the research and development to intelligentvideo surveillance to pay anyprice is worth.