Aiming at the problem of the low-frequency noise and drift error of the gyroscope that causes the accuracy of attitude measurement, this paper proposes to use the empirical wavelet transform (EWT) algorithm to fuse the gyroscope and accelerometer to calculate the attitude angle. First, use the EWT algorithm to divide the spectrum of the data collected by the gyroscope to obtain the modal components of the signal, and then use the wavelet adaptive soft threshold denoising method to denoise the signal and reconstruct the signal to obtain the processed gyroscope data. Then according to the PID complementary filtering method, the data of the accelerometer is used to correct the data of the gyroscope. Finally, use the corrected gyroscope data, combined with Runge-Kuta method to calculate the quaternion, so as to obtain the precise attitude angle by using the quaternion. The experimental results show that the EWT algorithm fused with gyroscope and accelerometer can improve the accuracy of attitude calculation by 50%, and the noise reduction effect is good, which meets the requirements of accuracy of attitude calculation.