A Gait Triaging Toolkit for Overlapping Acoustic Events in Indoor Home Environments

Area of Science:

• Gerontology
• Biomedical Engineering
• Machine Learning

Background:

• Gait analysis is crucial for assessing older adults' health.
• Acoustic gait detection offers non-intrusive data collection.
• Existing methods struggle in noisy home environments due to data limitations.

Purpose of the Study:

• To develop a machine learning filter for triaging acoustic gait samples.
• To improve the quality of gait datasets for training detection models.
• To enhance acoustic-based gait detection in real-world settings.

Main Methods:

• A novel machine learning filter was designed to identify and remove noisy gait audio.
• The filter prioritizes samples with clear spectral features and minimal noise.
• A deep learning model was trained with filtered and unfiltered datasets for comparison.

Main Results:

• The filter achieved an f(1) score of 0.85 in eliminating noisy gait samples.
• Gait detection performance improved by 25 points in f(1) score on unseen data when using filtered samples.
• The filter successfully enhances the quality of gait audio datasets.

Conclusions:

• The proposed filter significantly improves acoustic gait detection accuracy for older adults.
• Automating gait sample annotation will accelerate the development of healthcare monitoring tools.
• This approach is vital for reliable gait analysis in domestic and care settings.