Towards a Generalizable Method for Detecting Fluid Intake with Wrist-Mounted Sensors and Adaptive Segmentation

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Wearable Technology

Background:

• Mobile technologies enable health behavior monitoring.
• Automated dietary monitoring is researched, but fluid intake tracking is less explored.
• Passive sensing for health behaviors is a growing area.

Purpose of the Study:

• To develop and evaluate a method for passively detecting fluid intake gestures using inertial data from a wrist-worn device.
• To assess the generalizability of the proposed detection method across different participants.
• To contribute a labeled dataset for future research in fluid intake monitoring.

Main Methods:

• Adaptive segmentation of continuous inertial data streams.
• Utilizing a practical, off-the-shelf wrist-mounted sensor.
• Leave-one-participant-out (LOPO) cross-validation for performance evaluation.

Main Results:

• High precision (90.3%) and recall (91.0%) in detecting drinking episodes.
• Demonstrated generalizability of the fluid intake detection method.
• Successful recording of 561 drinking instances across 30 participants.

Conclusions:

• The proposed method effectively and passively detects fluid intake gestures.
• The approach shows strong generalizability, suitable for real-world health monitoring.
• The released dataset will foster further research in automated fluid intake analysis.