As Visual Background Extractor(ViBe) can not implement foreground detection precisely for a particular scene with dynamic backgrounds, This paper proposes a modified ViBe algorithm. It describes the original ViBe algorithm and its characteristics and discusses several modification schemes for the original ViBe in dynamic background scenes. Firstly, model initialization is conducted with several continuous frames instead of one single frame to handle ghosts. Then, self-adaptive threshold is adopted in the process of model matching so that background models is better suitable for the dynamic background. Finally, a spatial coherence estimation and a fuzzy rule in model maintenance are proposed to reduce false detections and to improve the robustness of the algorithm. Experiments demonstrate that the algorithm proposed detects effectively the movement targets in dynamic background scenes and its precision is improved by 20 percent as compared with that of the original ViBe algorithm.