Wearable Sensor Gait Analysis of Fall Detection using Attention Network.

Area of Science:

• Geriatric Medicine
• Biomedical Engineering
• Artificial Intelligence

Background:

• Falls are a major cause of mortality in senior adults, with one death occurring every 19 minutes in the US.
• Improving elderly care through early fall detection and protective measures like airbags is crucial.
• Wearable sensors and advanced algorithms are increasingly important for fall prediction using accelerometry data.

Purpose of the Study:

• To develop a fall prediction model for the geriatric population to mitigate fall injuries.
• To enhance the detection of impending fall risk for timely intervention.
• To improve the accuracy of differentiating between actual falls and near-fall events.

Main Methods:

• Application of a deep neural network (DNN) with attention mechanisms to analyze accelerometry data.
• Utilizing a transformer DNN with Time2Vec positional encoding and a Masked Transformer Network for gait analysis.
• Defining the observation window based on the maximum value of sensor signals.

Main Results:

• The proposed transformer attention network demonstrated superior performance compared to existing models.
• Achieved better specificity and sensitivity in fall detection using a custom dataset.
• Successfully modeled gait analysis for fall prediction.

Conclusions:

• The novel transformer attention network offers a promising approach for accurate fall detection in seniors.
• This technology can significantly contribute to reducing fall-related injuries and fatalities in the elderly.
• The model's enhanced accuracy provides a foundation for developing effective proactive fall mitigation strategies.