Abstract: In order to enhance vehicle video detection a vehicle segmentation method based on the fusion of adjacent frame difference and background difference in virtual window. The object area is ascertained by a new efficient background model update method. The region of non-target disturbance is reduced by corrosion and dilation morphological operations. Then inter-frame logic operations and seed filling method are used to fill the region to be divided. The vehicle can be detected by horizontal and vertical projection method in the end. Experiments show that this method has a good detection effects for road vehicles.