Moving Object Detection And Tracking For Video Surveillance

In recent years，with the rapid development of the performance of computerhardware and software, computer network technology and computer visiontechnology, Intelligent video surveillance which attracted the attention of scholars andresearchers, is already a hot emerging research question. Intelligent video surveillancesystem can reduce human interference, save resources and increase monitoringefficiency, which be widely used in transportation, banking, national security andother fields. The design of Moving target detection and tracking algorithm are the twocore technologies in the intelligent video surveillance system. Therefore, the study oftarget detection and tracking methods has important practical significance.In this paper, we consider a fixed camera capture video images as the researchobject to study target detection and tracking. The main work is as follows:(1) The thesis studies the moving object detection method, analyzes the workingprinciple of these methods, scope and the advantages and disadvantages, and alsointroduces details of several classical background modeling algorithms. Then thealgorithm for ViBe cannot be effective in detecting the presence of target because ofswinging tree branches background interference and other factors, the thesis improvethe fixed threshold of the algorithm which determined whether the pixel is thebackground pixel combining features of background model sample set. This caneffectively improve the detection rate and reduce the false detection rate. Theanticipated experimental results show this algorithm can get better detection effect.(2) The thesis introduces classification of target tracking algorithm and thecorresponding advantages and disadvantages and mainly described the basicprinciples and workflow particle filter tracking algorithms. On this basis, elementaryparticle filter algorithm in the scene of similar background color cannot be effectivelytrack the target. The thesis determines whether it encounter the interference of similarbackground by comparison of the relationship between the location which leastsquares method predicted and position of the particle filter algorithm estimated, and overcome the interference of a similar background, combined with the observationstatus information of objectives and the status information of objectives such as thedistance and direction of movement. So we can make up the traditional lack ofelementary particle filter target tracking algorithm. The experimental results showthat the algorithm is improved in accuracy and stability.