MEMS-IMU includes a tri-axial gyroscope and a tri-axial accelerometer, in which gyroscope's noises lead to MEMS-IMU's orientation inaccurate and result in external accelerations presenting errors. Based on the problem, a method of removing gyroscope's noises in real time is proposed. The third column of directioncosine matrix and gyroscope's bias are taken as state vector so that gyroscope's noises could be obtained online. Both the external accelerations and measurement noises of accelerometer are regarded as measurement residuals so as to measure the gravity vector in any trajectory. A Kalman filter is utilized to fuse the measurement results of gyroscope and accelerometer, in which the former results is used to estimate the state vector and the latter results is aimed to calibrate the state vector. The feasibility of this study is verified by comparing orientation and external accelerations when MEMS-IMU is in arbitrary quasi-static. Experimental results show the maximum errors of pitch, roll, yaw and external accelerations are 0.5°, 0.2°, 2° and 0.2 m/s² respectively, which are much better than that only based on gyroscope.