Research On Background Modeling Algorithms In Intelligent Video Surveillance System

Intelligent Video Surveillance is a hot research topic, which is theoretically based onthe technology of computer vision, digital image processing and pattern recognition, andtakes use of the strong computational ability of computers to process and analyze theimage data, so as to get the key information in the videos. Compared with the traditionalvideo surveillance, Intelligent Video Surveillance shows to be more powerful to deal withdata, more effective to monitor and more sensitive. Nowadays, Intelligent VideoSurveillance has been widely used in intelligent transportation systems, public securityand intelligent buildings.Background modeling, as a popular approach for moving objects detection, iscommon used in Intelligent Video Surveillance Systems to get the moving objects in thescenes. The result of background modeling will be provided for the higher levelapplications, such as people counting, objects tracking and so on. This paper mainly doesanalysis and research on the moving objects detection technology based on backgroundmodeling algorithms, and proposes two new kind of background modeling algorithms.Firstly, some research on three classical background modeling algorithms has beendone, Gaussian Mixture Modeling, Codebook and the Local Binary Pattern basedbackground modeling algorithm. Experimental results demonstrate that these algorithmscannot adapt to illumination changes and moving shadow in complex scenes well. Then,to deal with the problems of illumination and shadow, two new kind of backgroundmodeling algorithms have been proposed, the principle component analysis backgroundmodeling based on colorful bricks, and the background modeling based on Multi-ChannelScale Invariant Local Ternary Pattern(MC-SILTP).The first one adopts a new model, anddescribes the background with a background subspace. This approach can handle the tricky problems of illumination changes and moving shadows, and adapts to practicalscenes. The background modeling based on MC-SILTP introduces a newly developedtexture descriptor based on multi-channel, which can deal with the illumination changesand moving shadows, and adapt to dynamic scenes well.