Multi Phase Codebook Background Model And Application

Background model is the fundamental part of video application system. In order to generate background model rapidly and effectively, many frequently-used background models are analyzed, then present a novel model which named multi phase codebook background model.Multi phase codebook background model based on codebook background model, the expansion list blow:1.Frame-level Multi Phase Processing:During training and updating background model period, according to global average gray value of image, new codebook is generated, or the appropriate codebook is updated. During background subtraction period, the selective codebook is used. This allow our model to adapt diversity of background scene.2.Adding Weight for Codeword:Conception of weight is added into codeword in order to bring about the desired result. At first it can clear codewords which are too small at the right time. As result, the model becomes more performance and costs less memory. In the next place, weight can restrain noise and disturbance, it can also avoid foreground aperture problem. Moreover, weight can be used for acquiring the most reliable background image, Trenchant background image can be extracted from video sequence which contains confused moving objects.3.Shadow Detection:During background subtraction period, a optional shadow detection is chosen which based on physical characteristics of shadow. Shadow detection can eliminate shadow following moving objects, it also can wipe off highlight region around illuminant or light-reflecting objects.4.Inter-frame Processing:Optional region based process for foreground mask image include morphological open and close operation, contour based operation. This can eliminate noise and small foreground objects, conduce to plain and clean result.Using this method to analysis video sequence of trainman apartment, it can extract useful video segment include moving objects, and then save work time for manager who must check the video sequence.The results of experiments show that this new background model can avoid some common problems of past background models effectively. The model can be maintained fleetly and accurately, it also can accommodate complicated scene, and it has better performance than others.