Moving Target Classification Based On Image Sequence Classification

Identification of target classification based on video sequence images, is theimportant content to realize intelligent monitoring. It extracts the objects by motiontarget detection first, and then describes the movement of retrieved objects by feature,analysis its’ characteristics, as to the next moving object classification, and for a specificapplication environment accordingly. The realization of the intelligent video monitoringnot only save a lot of manpower material resources, but also brings more economicprofit, for the social and economic development of science and technology has played astrong boost.In this article, mainly studies moving object detection and recognition, motiontarget feature extraction and classification method, improvement and simulationanalysis, and other issues under the background of monocular fixed camera. And aroundit done the following several aspects work:1. On the side of detection of moving target, using the method of backgroundsubtraction based on gaussian mixture background model for moving object extraction,under the comprehensive comparison of moving target detection method existing.andthrough the basic mathematical morphology filtering operation to eliminate noise andtarget "empty", to get more accurate motion target. Finally, through the outlineanalysis to extract features of moving targets.2. In the aspect of target classification, by learning the basic principle of SVMand the basic training methods, and analysing the solution structure and space structureof the support vector, a SVM training sample reduction strategy which is on the basis ofcombining fuzzy vector machine (SVM) determine the membership degree method isimproved. By removing samples has nothing to do with the support vectors of supportvector, and the noise sample support vector counterproductive, in order to narrow thetraining sample set, to improve the SVM training speed.3. In view of the nonlinear characteristics due to lack of the concrete expression ofthe mapping changes in the space, resulting in class center cannot directly through thecomputed problem, used an idea that is looking for minimum super ball replacementclass is approximated by its center on alternative strategy in the feature space,simplifying the operation, reducing the time overhead.4. On the basis of above work, through the actual video data to the research on theeffectiveness of the simulation analysis. The simulation results show that it is effective and feasible.