In-Home Gait Abnormality Detection Through Footstep-Induced Floor Vibration Sensing and Person-Invariant Contrastive...

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Neurology

Background:

• Gait abnormality detection is vital for fall risk assessment and early diagnosis of neuromusculoskeletal disorders (e.g., Parkinson's, stroke).
• Current clinical gait assessments are infrequent and inaccessible, while existing in-home solutions have limitations (e.g., visual obstructions, limited coverage, device requirements).

Purpose of the Study:

• To introduce a novel, low-cost, non-intrusive, and device-free in-home system for gait abnormality detection using footstep-induced floor vibrations.
• To address the challenge of high uncertainty in floor vibrations due to inter-person gait variations and develop a generalizable model.

Main Methods:

• Analysis of time-frequency-domain features from floor vibration data during specific gait phases.
• Development of a contrastive learning-based feature transformation to create an embedding space robust to inter-person variations and sensitive to gait abnormalities.
• Utilizing a downstream classifier on transformed features for gait abnormality detection.

Main Results:

• A real-world walking experiment with 21 participants demonstrated the system's effectiveness.
• Achieved a mean accuracy of 85% to 95% in detecting various gait abnormalities.

Conclusions:

• The proposed method offers an accessible solution for in-home gait abnormality detection, overcoming limitations of previous approaches.
• This novel technique enables gait health monitoring without intrusive devices or the need for patient-specific labels for new individuals.