Algorithm On Video-based Dual Background Modeling

With the development of modern science and technology, more intelligent videosurveillance system is in deeply requirement,especially in airports,prisons,banks,museums,jewelry stores and some other occasions,the automatic alarm video surveillance has attractedmore and more researchers to study. Detection of Regional invasion Objects Based on Video isone of research which is most studied.Background modeling is the most used technology in the field. The mainly usedbackground modeling algorithms include moving average background modeling, Gaussianbackground modeling, Gaussian mixture background modeling and codebook backgroundmodeling algorithm. Each background modeling algorithm has its own advantages anddisadvantages.the effect will be different when it is applied to different video scenes andenvironments. Moving average background modeling is usually applied to the moving targetwhich the scenario is simple; The single Gaussian background modeling algorithm uses aniterative approach to establish the parameters of Gaussian model for each pixel in the imageframe of a single scene, in order to extract the former attractions; Gaussian mixture modelingis a improvements of single Gaussian model for more complex scenarios, the algorithm usesa number of features of the Gaussian model to represent each pixel in a image frame, eachframe of the video update the current model when it reads a new frame, and extract the formerattractions by matching pixels in the current frame and the background model; The codebookbuild its model code based on the pixel point in the video sequence of image frames, the codeapplied to the the case of complex video sequences of pixels of the image frame, thealgorithmbuild its model code according to the video sequence in the video, the model code links thecode words corresponding pixels in the frame, while reading each image frame pixel,themodel code of the information matches a code word, and whether matching the informationin the code word determines the final sequence of interest.In this paper, we propose a new dual background model that based on the research of thetraditional codebook and moving average background model. These algorithms are all basedon the application of the video sequence. This new background modeling algorithm proposesa updating method based on the HSV features and statistical methods, it filters out the noisepoints and solves the background updating problems under the illumination mutation effectively. Experiments show that this algorithm has a strong robustness to illuminationmutation, a fast response and it is quite fit for real-time image processing.The results show that the moving Average technology is faster than the codebook whenthe background environment is simple while the codebook is more clearly if it is on thecontrary.Experiments show that this Dual Background Modeling algorithm has a strongrobustness to illumination mutation, a fast response and it is quite fit for real-time imageprocessing.