Automated Detection of Gait Events and Travel Distance Using Waist-worn Accelerometers Across a Typical Range of Walking and Running Speeds

Albara Ah Ramli, Xin Liu, Kelly Berndt, Chen-Nee Chuah, Erica Goude, Lynea B. Kaethler, Amanda Lopez, Alina Nicorici, Corey Owens, David Rodriguez, Jane Wang, Daniel Aranki, Craig M. McDonald, Erik K. Henricson

Abstract

Accurately tracking gait clinical features (CFs)—such as step length, cadence, and total distance—is vital for monitoring the progression of neuromuscular diseases like Duchenne muscular dystrophy (DMD). Standard "black-box" algorithms in commercial pedometers often fail to detect steps at the slow walking speeds or irregular rhythms common in clinical populations. This paper introduces the Walk4Me system, a personalized, machine-learning-based approach to gait analysis using a single waist-worn accelerometer. By employing a brief five-speed calibration, the system achieves near-laboratory accuracy in measuring step counts and travel distance for children with DMD, significantly outperforming standard consumer pedometers.

Introduction

Monitoring gait progression in DMD is challenging because the disease causes stride lengths to shorten and vertical acceleration to decrease, leading to increasingly atypical patterns. While gait labs provide "ground truth" data, they are expensive and inaccessible for daily monitoring. Consumer devices, optimized for healthy adults, often exhibit error rates exceeding 80% in clinical contexts. Walk4Me seeks to bridge this gap by providing lab-grade precision through a wearable smartphone-based solution.

Methods: The Walk4Me Pipeline

The Walk4Me system utilizes the 3-axis accelerometer within a standard smartphone worn at the waist, near the body's center of mass. The core innovation is a multi-step, automated pipeline designed to handle the variability of atypical gaits.

The Walk4Me system architecture (B) and the personalized regression model (A) mapping acceleration to step length across different calibration levels (SC-L1 to SC-L5).

The system focuses on the anteroposterior (z-axis) signal, which is more reliable than the vertical axis for populations with reduced vertical movement. After applying a low-pass filter, machine learning identifies peaks to count steps and locates the "Initial Contact" (IC) point—the moment the heel hits the ground. Crucially, the system builds a personalized nonlinear regression model for each user, mapping the amplitude of anteroposterior acceleration to specific step lengths based on five short "speed calibration" walks (SC-L1 to SC-L5).

Experimental Evaluation

The study involved 30 participants: 15 children with DMD (ages 3–16) and 15 typically developing (TD) peers. Evaluation tasks included 10m, 25m, and 100m run/walk tests, the 6-minute walk test (6MWT), and unstructured "free-walk" (FW) evaluations.

Diagram of the 25-meter walking test path used to establish ground truth for distance and step metrics.

Ground truth was established via manual video observation by trained clinical evaluators. These standardized tasks allowed for a rigorous comparison between the Walk4Me model predictions and actual observed performance.

Key Results

The Walk4Me system demonstrated exceptional accuracy across both DMD and TD groups, maintaining stability across a wide range of speeds.

  • High Correlation: Pearson’s r values ranged from 0.9929 to 0.9986 for step counts and distance.
  • Low Error Rates: For the combined 6MWT, 100MRW, and Free-walk tasks, the mean percentage errors were 1.49% for step count and 1.18% for distance traveled. In contrast, the internal iPhone pedometer showed error rates of approximately 48% and 42%, respectively.
  • Step Length Precision: The mean percentage error for step length was remarkably low at 0.37%.

Discussion and Conclusion

The success of Walk4Me highlights two critical factors in clinical gait analysis: the advantage of the anteroposterior axis and the necessity of personalization. Universal algorithms cannot account for the "atypical" signatures of DMD, but a 5-minute personalized calibration enables high-fidelity tracking. By identifying the exact timing of Initial Contact, the system also allows for the calculation of advanced metrics like step duration and symmetry without expensive laboratory equipment.

Future work will focus on further automating the calibration process for remote setup and testing the system in uncontrolled "wild" community settings over longer durations.

Summary updated

BibTeX
@inproceedings{ramli2024automated,
  author = {Albara Ah Ramli and Xin Liu and Kelly Berndt and Chen-Nee Chuah and Erica Goude and Lynea B. Kaethler and Amanda Lopez and Alina Nicorici and Corey Owens and David Rodriguez and Jane Wang and Daniel Aranki and Craig M. McDonald and Erik K. Henricson},
  title = {{Automated Detection of Gait Events and Travel Distance Using Waist-worn Accelerometers Across a Typical Range of Walking and Running Speeds}},
  booktitle = {Sensors},
  year = {2024},
  url = {https://openreview.net/forum?id=9HYXk5CUBS},
}
Figure