Moving Object Detection And Shadow Elimination Based On Layered Codebook Model In Complicated And Dynamic Scenes

Background subtraction is the most common and effective method of change andmotion detection, and its key point is background modeling, background initializationand background maintenance. This paper made a lot of research based on the existingbackground modeling methods, and analyzed and summarized the development andapplication on this subject. This work gives us a point about further explore about theproblem of moving target detection based on background modeling under complexenvironment.In this paper, two classical methods Gaussian mixture model and ViBe arediscussed in detail with theoretical knowledge and implementation. The advantagesand disadvantages of each method are summarized with the experiments of randomnoise, irregular movement, camera jitter, dynamic scenes and shadows. The resultspoint out that these modeling methods do not completely solve the problem of motiondetection under these critical situations. In order to solve this problem, the improvingalgorithm layered codebook model to update the background in real-time is realizedbased on classical codebook model with more accurate test results. Comparative resultsindicate that the algorithm can handle scenes containing moving backgrounds orillumination variations, and it achieves robust object detection for different types ofvideos. Moreover, shadow severely affects the accuracy of moving objects detection,especially in complex environment conditions with a changed background. In thispaper, a algorithm based on layered codebook model and YCbCr color space isproposed to solve this problem. In the detection phase, the moving regions includingmoving objects and its shadow are extracted using layered modeling and detectionmethod accompanied with updating the background which consisted of the changedscenes at the same time. In order to detect the shadow against new background whichis obtained during the detection phase, these changes are used to update thebackground model. The background model is extracted from the permanent which obtained during the initial background modeling to avoid using static background inthe shadow detection applications. This allows us to have multiple layers ofbackground representing different background layers. After this, chromaticity feature isused to eliminate the moving shadow. We choose some videos from a public dataset,which is most commonly used in moving shadow detection and elimination, andexperimental results demonstrate that it is an adequate accuracy of the new adaptiveshadow detection and elimination algorithm and it is suitable for moving objectdetection under complex environment conditions.