The Design Of Intelligent Video Surveillance System In Traffic Intersection

Entering the21st century, the city’s public security has increasingly become people’sissues of concern. All of this are pressing for developing the urban security monitoringsystem. The way that traditional urban security monitoring system uses is artificialmonitoring, which will cause huge work to people as the number of urban surveillancecameras explode. Thus, the intelligent security monitoring system is imminent. In thispaper, we realize a intelligent security monitoring system without human interventionbased on video. Our system mainly realizes the functionality including real-time onlineabnormal monitoring and intelligent video retrieval. The main scheme of our systemincludes the motion detection, motion track, anomaly detection, etc. In the module ofmotion detection, the method that the paper uses is the improved GMM. As for the threeproblem in the traditional Gaussian mixture background modeling, including the sideeffect of a not so well initialization which will cause "ghost"、too much noise caused bythe original update strategy of variance、redundancy of the computation, this paper putsforward corresponding strategy, including self-adaption of update rate、threshold set ofvariance、 self-adaption of the K value. As for shadow removal, light correction andpost-processing in motion detection module, this paper also puts forward thecorresponding solutions. As for the module of motion track, this paper is based on theimproved classic blob tracking algorithm. We add a new corresponding way of blob whichthe original blob tracking algorithm does not consider. Also, for the issue of occlusion,including the static occlusion and the dynamic occlusion which the traditional blobtracking will fail, this paper puts forward the corresponding linear prediction and blocktracking solution. In the anomaly detection module, based on the prior exceptiondefinitions in combination with information including the location of the object, thetrajectory and motion state, our system can real-time online detect anomalies. For anomalywhich are not able to predict in advance, this paper puts forward a later video retrievalstrategy to supplement. Finally, through the system test and function test module it can beproved that the proposed intelligent video monitoring system can realize the functionwhich we set in advance, and has better real-time performance compared to other system.