A Real-World Dataset for detecting Handwashing in daily Life using Wrist Motion Data from Wearables

Area of Science:

• Activity recognition
• Wearable sensor data analysis
• Mental health technology

Background:

• Handwashing detection is crucial for hygiene and healthcare.
• Compulsive handwashing significantly impacts individuals with obsessive-compulsive disorder (OCD).
• Existing datasets lack real-world, unscripted compulsive handwashing data.

Purpose of the Study:

• To introduce the OCDetect Dataset, the first of its kind for compulsive handwashing.
• To enable the development of automated systems for real-time detection of compulsive handwashing.
• To support therapeutic interventions for individuals with OCD.

Main Methods:

• Collected data using inertial measurement units (IMUs) from 22 participants over 28 days.
• Recorded approximately 3000 instances of handwashing, annotated as compulsive or routine.
• Developed and evaluated a system for spotting handwashing from background activities on unseen participants.

Main Results:

• Achieved a maximum F1 score of 0.77 in detecting handwashing from background activities.
• The average F1 score was 0.33, significantly above the chance level of 0.03.
• The OCDetect Dataset provides a valuable resource for activity recognition research.

Conclusions:

• The OCDetect Dataset is the first to capture unscripted, in-the-wild compulsive handwashing.
• This dataset facilitates the creation of assistive technologies for OCD management.
• Publicly available dataset and code promote further research in activity recognition and mental health support.