An improved mixture-of-Gaussians model for background subtraction

The process of background subtraction is always a key step in surveillance. Currently, the mixture of Gaussians (MOG) works well in the background modeling and has been widely used in practice. In this paper, some new additional constrains are imposed on the updating process of statistics of Gaussian models. To reduce computational cost, the numbers of Gaussian models are selected dynamically based on the maximum recurrence time interval (MRTI). The experimental results show that the proposed method performs well in complex background modeling, and the efficiency in object detection is improved significantly.