A Novel Low-Rank and Sparse Decomposition Model and Its Application in Moving Objects Detection

At present, low-rank and sparse decomposition model has been widely used in the field of computer vision because of its excellent performance. However, the model still faces many challenges, such as being easily disturbed by dynamic background, failing to use prior information and heavy computational burden. To solve these problems, this paper proposes a novel low-rank and sparse decomposition model based on prior information, group sparsity, and nonconvex total variation. First, the rank of background matrix is fixed to 1, so singular value decomposition is no longer needed, which greatly reduces the computational burden. Secondly, the foreground target is divided into dynamic background and real foreground to reduce the interference of dynamic background. Finally, l(2,1)-norm and nonconvex total variation is introduced into model to incorporate prior information of dynamic background and real foreground. The experimental results show that compared with several classical models, our model can extract the foreground target from the dynamic background more accurately, more completely and more quickly.