Detecting Eating and Social Presence with All Day Wearable RGB-T.

Area of Science:

• Human-Computer Interaction
• Behavioral Science
• Wearable Technology

Background:

• Social presence influences eating behavior in individuals with obesity, but real-world monitoring is difficult.
• Existing high-resolution cameras have limitations in all-day data capture due to battery life and autonomy.
• Low-resolution infrared (IR) sensors show potential for automated behavior detection, but their combined use with RGB cameras is unexplored.

Purpose of the Study:

• To design and deploy a low-power system using RGB and IR sensors for detecting eating and social presence.
• To evaluate the effectiveness of a learned model in identifying these behaviors in real-world settings.
• To analyze system performance, device failures, and provide guidance for future wearable sensor studies.

Main Methods:

• Developed a low-power, low-resolution RGB camera and IR sensor system.
• Deployed the system with 10 participants with obesity for in-the-wild evaluation.
• Utilized a learned model to detect eating and social presence using combined sensor data.

Main Results:

• The combined RGB and IR sensor model showed a 5% improvement in detecting eating behavior.
• A significant 44% improvement was observed in detecting social presence compared to a video-only approach.
• Analysis of device failure scenarios provided insights for future wearable camera design.

Conclusions:

• Combining low-resolution RGB cameras with IR sensors enhances the detection of eating and social presence in real-world settings.
• This approach offers a promising, low-cost solution for unobtrusive, all-day human behavior monitoring.
• Findings guide future research in validating human behavior using context-aware wearable sensors.