On the Influence of Sample Rate, Calibration, and Allan Variance Parameters on the Accuracy of ZUPT-Based Pedestrian Navigation with MEMS IMUs

Abstract

MEMS IMUs are the most common units for data collection in the field of pedestrian inertial navigation, whereas on the data processing side a Kalman filter is a broadly used algorithm for the required pedestrian motion state estimation. The paper investigates the influence of several IMU-related parameters on the accuracy of this state estimation. It complements a former study of the authors that discussed algorithmic enhancements of the data processing leading to noteworthy accuracy improvements compared to the state of the art of pedestrian inertial navigation [1]. Therefore, the IMU-related parameters will be reassessed against this enhanced numerical background. The paper is based on test walks employing a special test track and a wearable overshoe equipped with two MEMS IMUs of different accuracy levels. The IMU data were then fused with a continuous-discrete total-state Kalman filter. The only aiding techniques were Zero Velocity Update (ZUPT) and Zero Angular Rate Update (ZARU) to exclude influences from other sensors than the IMUs. The paper considers the following points concerning their influence on the position and on the yaw estimation accuracy, which are especially error-prone motion states in pure ZUPT- and ZARU-aided inertial navigation: •Influence of the IMU sample rate for inertial data collection. •Influence of a pretest IMU calibration with respect to bias, scale factor, and sensor axes misalignment. •Incorporation of Allan variance sensor parameters in the Kalman filter. The parameters were used to model all inertial sensor biases by a random walk or a Gauss-Markov process of first order. The results on these points can be outlined as follows. •Higher sample rates show a significantly improved capture of the step dynamics leading to smoother and more accurate estimates of the walking path. •Against the enhanced numerical background mentioned above, the IMU calibration leads typically (but not always) to further accuracy improvements. •The state of the art in ZUPT-based pedestrian navigation does not show a clear advantage of using a Gauss-Markov process for the IMU sensor biases. This seems also to be the case for the algorithmic improvements mentioned above.