Enabling Automatic Monitoring of Fluid Intake and Medical Adherence by Human Activities Recognition from a...

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Gerontology

Background:

• Remote home monitoring is crucial for elderly and fragile individuals.
• Wearable sensors offer a non-invasive method for health status tracking.
• Activities of Daily Living (ADL) recognition is key for effective monitoring.

Purpose of the Study:

• To identify ADL using upper limb gestures from a single wrist-worn sensor.
• To specifically detect drinking and pill intake for medication adherence and hydration monitoring.
• To compare the performance of machine learning (ML) and deep learning (DL) models for ADL recognition.

Main Methods:

• Utilized a single Magnetic Inertial Measurement Unit (MIMU) placed on the wrist.
• Collected data from MIMU sensors (gyroscope and magnetometer).
• Trained and evaluated ML and DL models on three classification experiments for 14 ADL.

Main Results:

• Deep learning models outperformed ML models in distinguishing 14 ADL.
• The combination of gyroscope and magnetometer data achieved approximately 93% accuracy for general ADL.
• The same sensor configuration exceeded 97% accuracy in identifying drinking and pill intake.

Conclusions:

• Deep learning with MIMU data is effective for recognizing ADL, including critical indicators like drinking and pill intake.
• This technology enables continuous, minimally invasive monitoring of hydration and medication adherence in home environments.
• The findings support the development of robust solutions for elderly care and remote health management.