A Novel 3D Pedestrian Navigation Method for a Multiple Sensors-Based Foot-Mounted Inertial System

Wei Yang, Chundi Xiu, Jianmin Zhang, Dongkai Yang2017 · DOI

TL;DR

This paper introduces a novel 3D pedestrian navigation system that fuses data from a MEMS-IMU, barometer, and a permanent magnet. It addresses limitations in existing methods by improving stance phase detection for both walking and running, enhancing height estimation, and providing more accurate total distance calculations, leading to a robust and versatile indoor/outdoor navigation solution.

What Problem Does This Paper Address?

Accurate pedestrian navigation, especially in indoor or urban environments where GPS is unreliable, is a significant challenge. Foot-mounted inertial navigation systems (INS) offer an infrastructure-free solution but suffer from rapid sensor drift, leading to accumulated errors. The Zero-Velocity Update (ZUPT) algorithm is a common technique to mitigate this by using stationary stance phases to correct errors. However, traditional ZUPT methods struggle with accurate stance phase detection during running gaits due to their reliance on fixed thresholds. Furthermore, height estimation using barometers is susceptible to environmental fluctuations, and calculating the total traveled distance can be inaccurate due to high sensor sampling rates. This paper aims to tackle these issues by proposing a multi-sensor fusion approach with improved algorithms for stance detection, height estimation, and distance calculation.

Methods

The proposed system integrates three main components:

  • MEMS-IMU: Provides acceleration, angular rate, and magnetic field data.
  • Barometer: Used for height estimation.
  • Permanent Magnet: Attached to the foot to generate a magnetic field signal for improved stance detection.

The core algorithmic contributions are:

  • Magnet-based ZUPT Detector (Mag-ZUPT): Instead of relying on acceleration and angular rate thresholds, this method uses peaks in the magnetic field strength waveform from the permanent magnet to detect mid-stance phases. This approach is less sensitive to gait variations and works well for both walking and running.
  • Stance Phase Duration Model: Extends the mid-stance detection to a full stance phase by analyzing peak-peak periods in the magnetic field data to distinguish between walking and running gaits. This allows for a more accurate estimation of stance phase duration, which is crucial for improving ZUPT.
  • Height Difference Information Aided Barometer (HDIB): This algorithm improves height estimation by first using MEMS-IMU data to distinguish between level ground, upstairs, and downstairs movements. This information is then used to adaptively filter and correct barometer readings, making height estimation more robust to environmental noise.
  • ZUPT-based Adaptive Average Window Length (ZUPT-AAWL): To address the inaccuracy in total traveled distance calculation due to high IMU sampling rates, this method adaptively determines the window length for averaging trajectory data based on individual gait cycles. This ensures that the averaging window more closely matches the actual gait cycle, leading to a more accurate total distance measurement.

These components are integrated within an Extended Kalman Filter (EKF) framework for state estimation.

Datasets

The paper does not explicitly mention training datasets. Instead, it relies on experimental data collected from various scenarios to evaluate the proposed system. These scenarios include:

  • Walking, running, and mixed walking/running trajectories on a 100m runway.
  • Six typical trajectories involving walking on level ground, ascending/descending stairs, and indoor/outdoor environments.
  • Comprehensive pedestrian trajectory experiments covering complex indoor/outdoor paths with significant vertical movement.

Key Insights and Ideas

  • Magnetic Field Strength as a Robust Stance Indicator: The magnetic field strength from a permanent magnet provides a surprisingly stable and reliable signal for detecting mid-stance, outperforming traditional acceleration/angular rate-based methods, especially during running.
  • Gait Pattern Adaptation is Crucial: Stance phase duration and total distance calculations are highly dependent on the pedestrian's gait. A one-size-fits-all approach (like a fixed percentage for stance duration) fails in mixed or running gaits. Adaptive methods are essential.
  • IMU and Barometer Synergy for Height: While IMUs drift and barometers are noisy, combining them intelligently (as in HDIB) can leverage the short-term accuracy of IMUs to filter the barometer, leading to significantly improved height estimation.
  • Adaptive Windowing for Distance Accuracy: The high sampling rate of IMUs can lead to overestimation of distance. Adapting the averaging window length to individual gait cycles is a simple yet effective way to correct this.

Key Results

  • Mag-ZUPT Performance: In running scenarios, Mag-ZUPT correctly detected significantly more stance phases (e.g., 36 out of 37) compared to traditional ZUPT (e.g., 4 out of 37).
  • Total Traveled Distance Error (TTDE): TTDE ranged from 0.32% to 1.04% across various complex trajectories, demonstrating high accuracy. For example, Traj_1 achieved 0.32% TTDE.
  • Height Estimation Error: The HDIB algorithm achieved height estimation errors as low as 0 m and up to 2.35 m in complex scenarios, significantly outperforming standalone MEMS-IMU or barometer methods. For instance, in Traj_6, the height error was 2.35 m.
  • Overall System Accuracy: The integrated system demonstrated good 3D positioning accuracy in complex indoor/outdoor trajectories.

Limitations and Future Work

  • Low-Cost MEMS-IMU: The authors aim to achieve similar TTDE performance using lower-cost MEMS-IMUs in future work.
  • Stance Phase Duration Model: Further research is planned to develop more sophisticated models for stance phase duration and peak-peak period information, potentially using advanced machine learning techniques like artificial neural networks and random forests.
  • Experimental Scope: While the experiments cover a good range of scenarios, more extensive testing with diverse user populations and environments could further validate the system's robustness.

Summary updated

BibTeX
@misc{yang2017novel,
  author = {Wei Yang and Chundi Xiu and Jianmin Zhang and Dongkai Yang},
  title = {{A Novel 3D Pedestrian Navigation Method for a Multiple Sensors-Based Foot-Mounted Inertial System}},
  year = {2017},
  doi = {10.3390/s17112695},
}
Figure