A de-noising method is proposed for 3DM-GX1 accelerometer signal based on kalman filter and average filter. This method can reduce the disturbance caused by vibration of vehicle motion and sensor itself, get acceleration accurately, and have advantages of easy realization and good real-time performance. A vehicle acceleration signal model is established with real acceleration noises as simulation signals based on kalman filter and average filter. Experiments based on real vehicle acceleration signal showed that the method could reflect the detailed information effectively by filtering noise, and it is very suitable for de-noising of vehicle acceleration signals which needs real-time application. The start, stop, advance and retreat action could be identify from the acceleration signal by using the proposed de-noising method based on kalman filter and average filter.