Towards Environment-Aware Fall Risk Assessment: Classifying Walking Surface Conditions Using IMU-Based Gait Data and...

Area of Science:

• Biomedical Engineering
• Machine Learning
• Gerontology

Background:

• Fall risk assessment (FRA) is crucial for preventing falls in individuals.
• Wearable inertial measurement units (IMUs) enable free-living FRA but lack environmental context.
• Improving FRA with environmental awareness can enhance preventative strategies.

Purpose of the Study:

• To develop and analyze a convolutional neural network (CNN) model for classifying walking surface conditions using IMU gait data.
• To provide a foundation for environment-aware fall risk assessment (FRA) systems.
• To investigate the impact of gait signals, sensor placement, and segment size on classification performance.

Main Methods:

• A 25-layer CNN model was trained on IMU data from 30 participants walking on nine different surfaces.
• Data included acceleration, magnetic field, and rate of turn signals from six IMU sensors.
• Systematic analysis evaluated the influence of sensor type, placement, and data segment size.

Main Results:

• The CNN model achieved high accuracy in classifying walking surface conditions.
• Accuracies reached 0.935 with a single sensor and 0.969 with dual sensors.
• Optimal settings yielded a peak accuracy of 0.971, demonstrating robust performance.

Conclusions:

• The developed CNN model effectively classifies walking surfaces using IMU gait data.
• This approach enables the creation of more reliable and interpretable environment-aware FRA methods.
• Findings support the advancement of personalized fall prevention strategies by incorporating environmental context.