In the motion orientation estimation based on MEMS sensor technology, gyroscope signal drift error and gravity superimposed with linear acceleration are the two major reasons affecting the accuracy of estimation. In practice, static compensation and filter technology are commonly used to reduce the orientation estimation error. This paper designs a novel double stage extend Kalman Filter performed on self-developed inertial measurement unit. Above all, we construct adaptive acceleration error covariance matrix to eliminate the linear acceleration in quaternion-based orientation estimation model. Then, in order to correct the drift error produced by gyroscope, the multi-sensor data fusion technology is adopted to fuse the data. Experiment result indicates that the performance of our algorithm is in accordance with the motion capture system Xsens approbated widely. It proves that the algorithm can meet the application requirements effectively.