Wearable Sensor-Based Step Length Estimation During Overground Locomotion Using a Deep Convolutional Neural Network.

Area of Science:

• Biomechanics and Wearable Technology
• Machine Learning in Healthcare
• Gait Analysis and Rehabilitation

Background:

• Gait asymmetry poses significant health risks, including joint degeneration and impaired balance.
• Accurate step length measurement is vital for clinical interventions and gait training.
• Conventional foot-mounted sensors have limitations like signal drift and obtrusiveness.

Purpose of the Study:

• To develop and validate a deep convolutional neural network (CNN) model for step length estimation.
• To utilize proximal wearable sensors (hip goniometer, trunk IMU, thigh IMU) for gait analysis.
• To achieve accurate and generalizable step length estimation across various walking speeds.

Main Methods:

• A deep CNN model was designed for step length estimation.
• Data from sixteen able-bodied subjects were collected on a treadmill at varying speeds.
• The model was trained and validated using treadmill data and tested on overground walking data.

Main Results:

• The CNN model achieved an average mean absolute error of 2.89 ± 0.89 cm for step length estimation.
• The model demonstrated effectiveness across different walking speeds and conditions.
• Proximal sensor-based estimation proved viable and accurate.

Conclusions:

• The proposed CNN model offers a reliable method for real-time step length monitoring using proximal wearable sensors.
• This approach overcomes limitations of conventional methods, enhancing applicability in real-world scenarios.
• Findings support the integration of this technology into wearable assistive devices and personalized gait training programs.