3D Tracking via Shoe Sensing

Fangmin Li, Guo Liu, Jian Liu, Xiaochuang Chen, Xiaolin MaSensors 2016 · OpenReview

TL;DR

This paper introduces a novel 3D indoor tracking system that leverages low-cost inertial sensors embedded in shoes. By analyzing gait patterns and classifying walking states (walking, upstairs, downstairs), the system achieves accurate real-time 3D positioning, overcoming the limitations of GPS in indoor environments.

What Problem Does This Paper Address?

Global Positioning System (GPS) is the standard for location-based services, but it's ineffective indoors due to signal blockage. While various indoor localization solutions exist, many focus on 2D positioning and often struggle with accuracy, especially when dealing with the complexities of multi-floor environments. This limitation is critical in scenarios like emergency response or healthcare, where precise 3D location is vital. Existing inertial sensor-based solutions, often relying on mobile devices, suffer from accuracy issues due to the unpredictable movements of these devices. The paper aims to bridge this gap by developing a low-cost, accurate, and non-invasive 3D indoor positioning system using sensors attached to shoes.

Methods

The core of the proposed system lies in several key innovations:

  • Shoe-based Inertial Sensing: The system utilizes low-cost MEMS inertial sensors (accelerometer and gyroscope) attached to the user's shoes. This placement is found to be optimal for capturing stable walking patterns and gait information.
  • Gait Information Extraction: A "short-time energy-based approach" is employed to extract gait patterns. This method analyzes the energy fluctuation of acceleration signals, identifying moments when the foot is on the ground (low energy) as key gait events.
  • Walking State Classification: To improve vertical distance estimation, especially during stair climbing, a state classification model is developed. This model distinguishes between three states:
  • Plane motion (normal walking and jogging horizontally)

    • Walking upstairs
    • Walking downstairs This classification helps in accurately estimating vertical movement.
  • Error Reduction Mechanisms:

    • Posture Correction: Quaternions are used to represent and correct the sensor's orientation relative to a reference coordinate system. This involves initializing the posture using gravity vector and then iteratively updating it.
  • Cumulative Error Elimination: The system addresses the drift inherent in inertial sensors. It uses zero-velocity updates (ZUPT) during foot-ground contact to reset velocity and eliminate accumulated errors. A specific method is proposed to eliminate vertical distance errors by analyzing the "distance error gradient" and correcting the Z-axis movement.
  • Real-time 3D Trajectory Mapping: The system generates a dynamic 3D trajectory map using Unity3D, visualizing the user's movement in real-time.

Datasets

The paper does not explicitly mention pre-existing datasets for training. Instead, it relies on experimental data collected by the authors using their developed system. The experiments involve participants performing various activities (walking, jogging, going upstairs, going downstairs) in controlled environments (school corridors, stairs). The paper presents statistical analysis of this collected data to evaluate the system's performance.

Key Insights and Ideas

  • Shoe placement is crucial: Attaching sensors to the foot, rather than the thigh or leg, significantly improves the clarity and stability of gait pattern detection.
  • Short-time energy is a robust gait feature: This method proves effective in identifying steps, even across different walking states, outperforming simple acceleration magnitude-based methods for non-planar movements.
  • State classification is key for 3D accuracy: Differentiating between walking on a plane, going upstairs, and going downstairs is essential for accurately estimating vertical displacement and overall 3D positioning.
  • Low-cost sensors can achieve high accuracy: The study demonstrates that affordable MEMS sensors, when combined with sophisticated algorithms, can yield results comparable to or even better than more expensive systems in certain scenarios.

Key Results

  • Gait Information Extraction: Achieved nearly 100% accuracy in extracting gait patterns for walking and jogging using the short-time energy method.
  • Walking State Classification: The developed model achieved over 95% accuracy in distinguishing between walking/jogging, going upstairs, and going downstairs.
  • Vertical Distance Error Elimination:
    • While walking: Maximum error 0.291 m, average absolute error 0.1186 m.
  • While running: Maximum error 0.66 m, average absolute error 0.17 m. The system effectively reduced vertical distance errors.

  • Step Length Estimation:

    • Normal walking: Accuracy of 90.83%.
  • Jogging: Accuracy of 91.87%.
  • Going upstairs (horizontal): Accuracy of 90.83%.
  • Going upstairs (vertical): Accuracy of 87.5%.
  • Going downstairs (horizontal): Accuracy of 80.0%.
  • Going downstairs (vertical): Accuracy of 79.1%.
  • Heading Verification:
    • Walking: Mean course error close to 0°, maximum yaw angle error 15.46°.
  • Jogging: Maximum error 39.13°, average absolute heading angle error 7.09°.
  • Comparison with Literature: The system's step estimation accuracy was found to be close to a high-precision fiber optic gyro-based system [8], even outperforming it in the case of jogging, despite using significantly lower-cost sensors.

Limitations and Future Work

  • Limitations:
    • The paper acknowledges that while the system achieves high accuracy, the accuracy of step length estimation for downstairs movement is lower (around 79-80%) compared to other states.
  • The comparison with a high-precision system highlights a difference in sensor noise (22.5-fold), suggesting potential for further improvement with higher-quality sensors.
  • Future Work:
    • The authors suggest that large-scale deployment of shoe sensing systems would require careful consideration of networking metrics like throughput, delay, and energy efficiency. They leave the study of large-scale shoe sensing deployment for future work.
  • The paper implicitly suggests that further refinement of the algorithms, particularly for downstairs movement, could improve overall accuracy.

Summary updated

BibTeX
@inproceedings{li2016d,
  author = {Fangmin Li and Guo Liu and Jian Liu and Xiaochuang Chen and Xiaolin Ma},
  title = {{3D Tracking via Shoe Sensing}},
  booktitle = {Sensors},
  year = {2016},
  url = {https://openreview.net/forum?id=J0H45lzVnO},
}
Figure