Aiming at the problem of mutual interference between attitude algorithm and external force acceleration estimation in inertial navigation system, a new method based on quaternion and extended Kalman filter was proposed. Firstly, the acceleration data of the sensor was corrected by using the estimated external force acceleration data to obtain the accurate reverse gravity acceleration, combined with geomagnetic field vector and calculated by the gradient descent algorithm, the rotate quaternions were obtained. Secondly, the extended Kalman filter model was constructed to update the rotate quaternions and external force acceleration, the prediction value of rotate quaternions and the external force were obtained. Finally, the measured values of rotate quaternions and the acceleration data were corrected by Kalman filtering method, the accurate rotate quaternions and the external force acceleration of the three axis directions in reference coordinate system were obtained. The experimental results show that the method for the synchronization estimation of attitude and external force acceleration by extended Calman filter can quickly converge and accurately get the information of the attitude and the external force acceleration, its Euler angle error is ±1.95° and acceleration error is ±0.12 m/s ². The method can effectively restrain the influence of the external force acceleration on the attitude algorithm, and accurately estimate the external force.