The Research On Moving Objects Detection Methods In Intelligent Video Surveillance System

Video surveillance is widely used in the field of security monitoring system,thereare several advantages for using video surveillance,such as vision-illustrative, accurateand timely, rich contents and so on. And moving objects detection is one of the mostimportant core technology in the transition from traditional video surveillance tointelligent video surveillance system. Moving objects detection is the basis of objectstracking, objects classification, and behavior understanding of moving objects, for weonly take moving objects regions into consideration. Effective moving objectsdetection in complex environment such as changing illumination, moving shadows,changing scenes is one of the hottest topics in the field of computer vision. In thispaper, the methods for finding moving objects in complex enviroment under videosurveillance are discussed, the main contents of this paper are as follows:Firstly, three basic moving objects detection algorithms including simplebackground subtraction, consecutive frames difference and running average areanalysed and compared. The classic OTSU method is used to find the optimalthreshold, then the background subtraction and consecutive frames difference arecombined to improve the result of moving objects detection. The improved methodhave better performance but with higher computational cost and lower practicality. Inorder to modify the binary result frame, the morphology filtering operation andconnected components analysis based on contour are used.Secondly, the traditional mixture gaussian model is detailed discussed includingthe theroy of modeling and how to update the model. The notion of frame changingfactor is proposed for the purpose of selecting different learning rate so that the modelcan adapt the quickly illumination change and scene change. Then a improved mixturegaussian model is addressed, the new algorithm can automatically select the needednumber of components per pixel and in this way fully adapt to the observed scene.Moving shadows cause serious problems while segmenting and extracting movingobjects due to the misclassification of shadow points as foreground. Special attentionis paid to the suppression of moving shadows in the HSV color space and RGB colorspace. The improved mixture gaussian model algorithm is used to discriminate themoving object pixels and moving shadow pixels.Thirdly, a new method for finding moving objects by using random strategy isdiscussed. A unique decision rule is established to discriminate the moving pixels and the background pixels and the background model is initialized by the first frame of thevideo sequence. When updating the background model, not only update the samplesequence of the current pixel, but also update the neighborhood’s sample sequence.The spectral,spatial and the temporal features of pixels are exploited. The proposedmethod, compared with the typical methods, is proved to be simple but effcient withlower computational cost and higher accuracy.Fourthly, a simple moving objects detection system for video surveillance isdesigned. The video capture part is developed based on Directshow. In order to makeuse of OpenCV, the capture frame should be converted to IplImage structure so that thedata of capture frame can be analysised by the powerful functions in the library. Thenthe random strategy method is utilized for finding moving objects in the capturesequence, and connected components analysis is used to count the number of movingobjects and to locate the center position of moving objects. In the end, the explicitimplementation steps to design the interface of the system based on MFC areexplained.