This paper presents a method for vehicle flow detection statistic at traffic crossing based on computer vision and image processing. It can affectively use optical flow to detect traffic parameter by analyzing vehicle and scene information extracted from images. The experimental results show that this method is correct and efficient.