Research On Designing And Development In Intelligent Video Analysis System

With the widely application of the video surveillance system, video surveillancebecomes an important technical means of security and defense for the state organs,enterprises and institutions, as well as private property. Traditional video monitoringmainly depends on artificial monitor, so it is difficult to avoid the false alarms andmissed alarms that caused by fatigue or absence of the staff. With the improvement ofcomputer performance and the development of image processing technology, theintelligent video surveillance demonstrates many good features, such as highlyself-motivated alarm, long time work and low cost manpower. Thus it becomes thetrend of the development in video surveillance system. As the core of the intelligentvideo surveillance, the research of intelligent video analysis is of great academic valueand broad market application.In this paper, the technology of intelligent video analysis is studied, and some basicfunctional modules are developed, such as intrusion detection, direction detection,tripwire detection and fire detection. Furthermore, an intelligent video surveillancesystem for mass rapid transit and an all-weather cooperative tracking system aredeveloped. The algorithm theory and realization about the function module and theapplication system is described in detail from the following aspects, including theproject design, key technology research, system building and running effects.The research and development work carried out in this paper can be summarized asthe following aspects.On the basis of the study of traditional video surveillance technology and imageprocessing technology, this paper designed an intelligent video analysis system, whichhas four common functions including intrusion detection, direction detection, tripwiredetection and fire detection. Intrusion detection improves the modeling stage andupdating stage compared with the standard method of frame difference algorithm. Itmakes the background model better to be able to adapt to the changes in theenvironment of the monitoring field-of-view, which reduces the large number of falsepositives and false negatives. Direction detection and tripwire detection effectively useof the results of the processing of the intrusion detection. By obtaining the inter framerelations association of the prospects moving target and the forecast of movement trends,these two modules can detect the behavior of wrong direction or crossing the tripwire. Based on the statistical characteristics of fireworks in the YCrCb color space combiningwith the policy decision fuzzy inference system (FIS), we developed the fire detectionfunction. Compared to the conventional fire detection method, our method can greatlyreduce false alarms caused by red object and the light to achieve real-time accuratelyalarm of fire. What’s more, the author integrated and packaged the above-mentionedmodules, and developed a software development kit named IVAT(Intelligent VideoAnalysis Toolkit), which is lightweight and intelligent. It has been successfully appliedto the Smart Security Eyes intelligent video analysis system developed by ourlaboratory.This paper presents an intelligent intrusion detection program using multispectralimaging for the monitoring of long-distance rail lines. Whether the weather and lightchanges, uncooled focal plane infrared thermal imagers can provide high quality videos.And, a visual field perspective reconstruction technology is proposed to effectivelyeliminate the perspective phenomenon, which makes the target subject to a unifiedrange of the threshold constraint in the different position of the rail line. Reducing falsealarms and missed alarms leaded by perspective phenomena greatly, this system caneffectively protect long-distance rail lines. In addition, based on the introduction ofmultilevel alarm regions, this system can assess the degree of threat for the protectedarea. The program has been successfully applied to the Yu’ao Bridge section ofChongqing Metro Line now.The paper proposes an all-weather videos collaborative target tracking system forthe main roads protection of the organs and communities. Using the same imagingapparatus mentioned above, the function of background segment becomes robust PTZcameras track the moving target stably employing a particle filter target trackingalgorithm based on the target partition described. Furthermore, the paper uses alightweight local feature registration method for the matching between the field of viewin different cameras.