Learning-based stance-phase detection for a pedestrian dead-reckoning system with dynamic gait speeds. (21st June 2021)

• Record Type: Journal Article
• Authors: Zhang, Liqiang; Wu, Jianchao; Jiang, Chunxu; Jing, Peiguang; Liu, Yu
• Is Part Of: Measurement science & technology. Volume 32:Number 10(2021)
• Publication Date: 2021-06-21
• Subjects: adaptive thresholding -- dynamic gait speeds -- pedestrian navigation and robust stance phase detection; Physical measurements -- Periodicals; Scientific apparatus and instruments -- Periodicals; Equipment and Supplies -- Periodicals; Science -- instrumentation -- Periodicals; Technology -- instrumentation -- Periodicals; Mesures physiques -- Périodiques; Physical measurements; Scientific apparatus and instruments; Periodicals; 502.87
• DOI: 10.1088/1361-6501/ac0072
• Languages: English
• ISSNs: 0957-0233

Abstract

Abstract: Due to the errors arising from low-cost inertial measurement units, pedestrian dead-reckoning (PDR) systems suffer from accumulative velocity and position errors. Previous works proved that the zero-velocity update (ZUPT) method can effectively reduce the velocity errors of the PDR system so that positional errors can be mitigated. The ZUPT algorithm is triggered by a stance phase-detection module that generally adopts a fixed threshold. However, the fixed-threshold-based detector is not robust enough and may not recognize stance phases accurately with dynamic gait speeds. Incorrect detection degrades the positioning accuracy of the overall system. In this paper, to improve the performance of the stance-phase detector, we explore the relationship between threshold and gait speed, and then threshold regression problems are constructed to design machine-learning-based stance-phase detectors that are robust to gait speed. Real-world highly dynamic experiments have illustrated the effectiveness of the proposed methods with dynamic gait speeds. Compared to the best fixed-threshold-based traditional method, the experimental results show that the machine-learning-based methods reduce the minimum root mean squared error (RMSE) of the distance measurement by 22.5%–37.2%, the minimal RMSE of the start–end error by 5.8%–13.2% and the average RMSE of the positional error by 12.4%–17.5%.