Ghosts and stationary foreground detection by dual-direction background modeling

Chaste and stationary foreground may occur in traditional background subtraction when objects start or stop moving. Eliminating ghosts and extracting stationary foreground immediately are crucial for improving the subsequent tasks such as object tracking, recognition and activity analysis. In this paper, we propose a method to detect ghosts and stationary foreground by dual-direction background modeling. The forward background model and the backward background model are built by GMM and a simple regression model respectively, which can detect not only the moving foreground but also the stationary foreground and the ghosts. Extensive experiment results demonstrate that the proposed algorithm is effective and efficient in eliminating ghosts and detecting stationary foreground.