Aiming at the problems that the visual background extraction model is difficult to adapt to dynamic scenes, appears the target holes easily, spends much time to eliminate ghosts, and pixels at the junction of the foreground and the background are prone to retransmit wrong information. A video region of interest extraction algorithm based on visual background model improved is proposed. In the foreground segmentation phase, according to the spatio-temporal correlation, the global quantity threshold is adjusted by using the mean absolute deviation adaptively, and a dynamic quantity threshold is obtained. At the same time, according to the contribution rates of the scene regions to the human eyes, the variance is used to adjust the global distance threshold adaptively, and the dynamic distance threshold is obtained, and the propagation probability of misdetecting information is eliminated and the detection accuracy is improved, which adapts to the changes of dynamic scene, and the video region of interest is extracted. In the update phase of background model, the complexity of regional scene is used to adjust the update periods dynamically to eliminate the target holes and accelerate the elimination of ghosts effectively. The results show that the proposed algorithm can make up the shortcomings of the VIBE algorithm, and extract the region of interest of video with higher precision, which is suitable for the extraction of the region of interest in the dynamic video scenes.