A Novel Drift-Robust Method for Position Estimation Based on Linear Kalman Filter

Position estimation is a key challenge in many applications where inertial measurement unit (IMU) sensors are commonly used for this purpose. Double integrating the experimental acceleration measured by IMUs may result in a signal that grows quadratically with time. This effect in the position estimation process is called drift. The drift error is also arisen in the measurement of intermittent body movements such as walking or hand tremor. In this paper, we propose an innovative approach to solve this error based on the traditional linear Kalman filter. This idea is according to the implicit involvement of the assumption of periodicity of the motion in the state equations and the use of Fourier expansion, which leads to the position estimation being more robust to drift. We evaluate the accuracy of the proposed idea via evaluating its performance on some periodic signals as well as the experimental signal obtained from hand tremors. The results indicate that this method is robust to drift without additional computational cost, and it performs superior in terms of consistency compared to the traditional Kalman filter parameters.