Automated Fluid Intake Detection Using RGB Videos.

Area of Science:

• Gerontology
• Computer Science
• Artificial Intelligence

Background:

• Dehydration is a prevalent and serious health concern for older adults.
• Forgetting to drink regularly contributes to dehydration and associated complications in this demographic.
• Existing automated detection methods often rely on static images and controlled lab settings.

Purpose of the Study:

• To develop and evaluate an automated system for detecting drinking events in older adults.
• To address the limitations of current vision-based methods by utilizing video segments.
• To compare the performance of 3D convolutional neural networks (CNNs) against 2D CNNs for drinking event detection.

Main Methods:

• A 3D CNN model was proposed for detecting drinking events using video segments.
• Data was collected from 9 participants in a simulated home environment, including daily activities and drinking from various containers.
• The 3D CNN was trained and compared against a 2D CNN using both static images and video data.

Main Results:

• The 3D CNN model demonstrated superior performance compared to the 2D CNN.
• The 3D model achieved an F1 score of 93.7% with 10-fold cross-validation.
• An F1 score of 84.2% was achieved using leave-one-subject-out cross-validation with the 3D model.

Conclusions:

• 3D CNNs using video segments are effective for automated drinking event detection.
• This approach offers a more robust solution for monitoring fluid intake in real-world settings.
• The findings support the potential of AI-driven tools to manage dehydration risks in older adults.