Autonomous Pedestrian Altitude Estimation Inside a Multi-Story Building Assisted by Motion Recognition

Abstract

Owing to autonomy and continuity, a pedestrian navigation system (PNS) has been widely deployed, which is based on the micro electro-mechanical system inertial measurement unit (MEMS-IMU) and the strapdown inertial navigation system (SINS). However, altitude information cannot be effectively obtained in this system by the double integral of vertical acceleration because of the altitude channel divergent of SINS. The study is aimed at improving the accuracy and robustness of altitude estimation through a novel method based on foot-mounted MEMS-IMU. More specifically, the proposed method exploits the adaptive network-based fuzzy inference system (ANFIS) to recognize vertical motion modes including horizontal, downstairs, and upstairs movements. Then, the pseudo height model based on both motion modes and the stair height is constructed for stair walking with different height. Finally, the pseudo measurements from the pseudo height model are integrated with the data from motion prediction through the extended Kalman filter (EKF). Experimental results show that the overall classification accuracy of ANFIS can reach up to 99.1%. Since ANFIS is utilized to assist height estimation, the cumulative height error accounts for about 1.2% over a total height of 44 m when a pedestrian walks up and down six floors without external facility and barometric pressure support. It is concluded that the ANFIS-based height estimation method can achieve better vertical positioning performance for PNS than the existing approaches in terms of accuracy and robustness.