Gait Event Detection and Gait Parameter Estimation from a Single Waist-Worn IMU Sensor

Area of Science:

• Biomechanics
• Wearable Technology
• Machine Learning

Background:

• Gait changes are linked to fall risk and neurological disorders.
• Inertial Measurement Unit (IMU) sensors offer continuous gait monitoring.
• Waist-worn IMUs are convenient for gait analysis, usable via belts or smartphones.

Purpose of the Study:

• To evaluate the accuracy of estimating gait events and parameters using waist-worn IMU data.
• To compare two machine learning (ML) sequence-to-sequence (Seq2Seq) models for gait analysis.
• To validate ML model performance against the GAITRite® system.

Main Methods:

• Utilized data from 69 subjects, encompassing 17,643 steps and 3588 walks.
• Implemented and compared two ML-based Seq2Seq algorithms: Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM).
• Assessed gait event detection (heel strike, toe-off) and step length estimation accuracy.

Main Results:

• The CNN-based algorithm demonstrated superior performance over the LSTM method.
• Achieved 98.94% accuracy for heel strike detection and 98.65% for toe-off detection.
• Obtained a mean error of 0.09 ± 4.69 cm for step length estimation.

Conclusions:

• Waist-worn IMU sensors combined with CNN-based ML models provide accurate gait event and parameter estimation.
• This technology holds potential for non-invasive monitoring of gait dynamics and fall risk assessment.
• The findings support the use of wearable sensors for clinical gait analysis and early detection of movement disorders.