This paper propose that with the application of the Kalman filtering technique, the leveling process of inertial navigation system with base motion can be carried out in a much short time. The base motion is generated by a second order Markov process. Furhter, for a long time huge disturbance during the vehicle motion, using the input estimation technique to modify the Kalman gain computation can obtain a good leveling error estimated. The results can be integrated into a true motion to provide a fast accuracy method for initial leveling.