Research On Background Subtraction In Intelligent Video Surveillance

Background Subtraction(BS, for short) aims at extracting moving object from complex and changeable scenes. As a classical but significant technology of computer vision research, it has ben widely applied in intelligent video surveillance. In dynamic scenes, BS tries to set up the foreground and background model to fit the distribution by analyzing the modal changes of background and foreground. In general, background modeling approach has the assumption that the scenes is simple, which is obviously not conforms the actual video surveillance. Illumination change, busy background and shadow always exist and will result in inaccuracy of background subtraction. Therefore, the research of the BS method has the theoretical and practical significance.Based on the probability density model and low rank representation, this paper goes deeply into the subject of BS algorithm in intelligent video surveillance and the main contribution is as follows:1. We lucubrate some common BS method and come up with an indeterminate coefficient GMM. By the experimental simulating, we compare the new algorithm with others.2. The existing ViBe mechanism has been studied and on the basis of analysis the burden of it, we propose an enhanced ViBe. The new idea makes the algorithm more reliable when faced with shadow and illumination changes by optimizing the adjacent sampling.3. The BS based on the low rank representation is deeply studied. In order to accurately separate foreground objects and reduce the noise effects as much as possible, we introduce the latent data. The result of experiment shows that our method can rarely manage the illumination change and noise.