As the existing Kalman Filter used in vehicles detection had some disadvantages of fixed background updating parameters and poor background modeling real-timeliness, proposed an vehicle detection algorithm based on three-frame-difference and adaptive Kalman filter method. Firstly extracted the vehicle motion region by three-frame-difference method, then determined the background updating parameters by Gaussian distribution and updated the background model, lastly obtained final result by the two subtracting image. Experimental result shows that the improved algorithm updates backgrounds quickly and extracts moving targets effectively.