Background modeling for dynamic scenes using tensor decomposition

Background modeling is an important topic in video analysis and understanding, while the main difficulty lies in the dynamic scenes. The scenes with limited variation can be effectively described by numerous models, but more precise representation model are needed for complex scenes. In this paper, we propose a background modeling algorithm for dynamic scenes using tensor decomposition. Since video data can be naturally represent as higher-order tensors, the tensor methods can accurately describe the dynamic nature of the nearby pixels, and then background will be estimated by the low-rank representation of tensors. Experimental results show that the proposed method is robust and adaptive in dynamic environments, and furthermore, moving objects can be clearly separated from the complex dynamic background.