Research On Moving People Detection And Tracking Algorithm In Video Surveillance System

With the development of concept of safe city, a large amount of cameras aredeployed, especially in public places that have great flow density and heavy securitytasks, to implement surveillance and assist in security. The traditional manualmonitoring, which takes too much manpower, material resources, has been graduallyreplaced by intelligent surveillance. Object detection and tracking technologies inintelligent video surveillance system, used in public places, can catch the personnelmovement real-timely, and therefore improve public safety.In object detection section, the deficiencies of the traditional frame difference andbackground subtraction when in public places were analyzed. Due to the deficiencies,foreground region was detected by correlation coefficient of two sequential frames anda background updating strategy was designed. Firstly, a square window was created. Bycalculating the correlation coefficient of pixel values in two images within the windowarea, the pixel could be judged whether it was the foreground pixel according athreshold. The method of correlation coefficient could resolve noise interference muchbetter and relieve the influence of non-stationary background. Secondly, the backgroundmodel can be updated by using brightness gain mean of pixels with the same gray levelin images. The last was to introduce the motion estimation mechanism to reduceinfluence of light changes. The experiments show that this method can suppress noisemuch better and accurately extract the foreground. Comparing with the Gaussianmixture model, the background model proposed in this paper can be better to reduce theimpacts of light changes.In the object tracking section, Camshift tracking algorithm based on the colorhistogram was adopted for the non-rigid feature of the human body during movement,such as scale changes. But Camshift algorithm had two shortcomings when in practicalapplications.â‘ The object must be labeled manually;â‘¡It was susceptible to theinterference of background with the similar color. Therefore, some improvements weremade as follows: by combining the above object detection algorithm, the object can betagged automatically; the object color modeling process of Camshift was improved byusing weighted color histogram model to reduce the interference caused by the similarcolor background for tracking. When the characteristics information of object was reduced because of sheltering, the object could be steadily tracked by employingKalman filter. The experiments show that the improved algorithms not only satisfy thereal-time requirement but also effectively track the object.Finally, based on the OpenCV, an intelligent video surveillance system wasimplemented, a system that, using the C/S architecture, can be applied to the actualmonitoring sites with easy deployment.