Research On Moving Objects Detection In Intelligent Video Surveillance System

As the newest generation of intelligent management and monitoring system, in theabsence of supervision, Intelligent Video Surveillance (IVS) can identify the anomaliesand real-time alarm through intelligently and automatically analyzing the video contentcaptured by surveillance cameras. With the rapid development and widely used of the IVS,as the key technology, moving target detection is also increasingly being interested,especially the effective detection under the complex scenes, such as climate change,illumination change, scene change and the background shaking, has become a hot researchin the field of IVS in recent years.In this paper, facing to the existing problems in the real scene, we mainly study theapproach of modeling background pixel-based and regional-based respectively, and twosolutions of modeling and updating background based on subtraction of background areproposed referring to the monitoring situation including background shaking and scenechange. In the first approach, brightness and hue in Lab color space are selected as theclassification feature, the OTSU algorithm is used to adaptive threshold classify, andupdate the background according to the advantages of background difference algorithmand frame difference algorithm. Additionally, in order to exclude the impact of messynoise in the process of background updating, the multi-frame statistics based jointjudgment is embedded in the updating algorithm for an uncertain region. In the secondmethod, except for the feature of single pixel itself, the correlation between adjacent pixelsis also fully considered. In order to reducing the complexity of describing the texturerelevance and making the weight of neighboring pixel to be in correspondence with thetarget pixel’s, the improved LBP equivalence mode is used, then the approach of modelingbackground combined with color feature and local texture information is proposed. To verify the performance of the proposed approaches, a lot of comparativeexperiments were done, especially for the second scenario, two indicators includingdetection rate and false detection rate are calculated for quantitative analyzing.Experimental results show that the algorithm in this paper have good performance whenthe monitoring situation includes background shaking and scene change.