Aiming at the problem that vehicle vibration affects the heading accuracy of MEMS IMU, a method that can effectively suppress vibration noise and improve heading accuracy and stability is proposed. Firstly, the minimum mean method is used to preprocess the data to improve the signal-to-noise ratio. Then, the bias noise of gyroscope is filtered by using the complementary characteristics of accelerometer and gyroscope. Finally, the extended Kalman filter is used for further filtering. A total of 4 hours of field experiment results show that IMU is less affected by carrier vibration, and the accuracy and stability of heading are improved. The relative heading error after large-angle mechanical motion is 3.08 ° and the heading variance at rest is 2.44＊10-5.