Abstract: Track irregularity is the main excitation source of the vibration of railway vehicle system, and the change of the running posture of the car body frame will affect the detection accuracy of track geometric irregularity. In theory, the running posture of the car body frame can be measured by the displacement sensor, but in most cases, the measurement is difficult and the accuracy is low. When the acceleration sensor is used to test, there are various kinds of noise interference, so that the integral displacement contains the trend term drift error. To solve this problem, a low-frequency attenuation frequency-domain integration method was proposed in this paper to obtain displacement time series from acceleration data. The method started from the integration of acceleration measurement, and used the least square method based on moving average to remove the error caused by random noise in the integration process and restored the actual vibration response caused by track irregularity. Combined with the moving average method, high pass filter and frequency domain integration, the reconstructed displacement signal was obtained through multiple rounds of smoothing, filtering, and integration to avoid signal distortion. The main advantage of this method was that there was no long period drift in the reconstructed displacement. The accuracy and reliability of this method were verified by comparing with the measured data of shaking table.