Robust Error State FIR Filter and Its Application in Pedestrian Tracking
Wenchen Liu,Yuan Xu,Zhiyun Yin,Jing Liu,Rui Gao,Zhongwei Shi,Kun Liang,Yan Huang
DOI: https://doi.org/10.1109/access.2024.3420098
IF: 3.9
2024-01-01
IEEE Access
Abstract:In recent years, pedestrian tracking algorithms have witnessed extensive utilization in various domains. Particularly noteworthy is the attention garnered by hospital pedestrian tracking due to its potential to enhance operational efficiency and patient’s safety. The continuously stable and accurate patients’ position information can help hospitals to guarantee that patients remain within designated areas and get timely assistance when necessary. In this work, in order to improve the robustness of the data fusion filter for the pedestrian tracking, one robust error state FIR filter (ESFIR) will been proposed. To the data fusion model, the quaternion-based data fusion models for the accelerometer will be firstly derived, which can improve the accuracy of sensor data for the Zero velocity UPdaTe (ZUPT) by using the quaternion-based state vector and the accelerometer value as the measurement. Then, two filters are used to reduce the error accumulation in inertial navigation system (INS): with the data of accelerometer and gyroscope, the ESFIR filter is employed to provide the error estimation of the accelerometer, which is used to correct the accelerometer’s output directly, then, the finite impulse response (FIR) filter will be used in the Zero velocity UPdaTe (ZUPT) method. With the Gyroscope’s output and the ESFIR corrected accelerometer data, the human position can be computed. Finally, four real tests will be done to verify the performance of the proposed method. The actual test results show that the accuracy of the position information derived from the solution using this method is improved by 8.43% on average compared with the conventional Kalman-based method when the setting is friendly to KF. When the setting is bad for KF, the KF-based method has divergence, and the proposed method can still provide stable solution, it can improve the localization accuracy by about 11.81% when compared with the INS+ZUPT method, thus providing better robustness.