The SPLENDID chewing detection challenge.

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Nutritional Science

Background:

• Wearable technology advances enable new methods for monitoring eating behavior.
• Chewing activity and eating occurrences are key aspects of eating behavior.
• Existing research lacks publicly available datasets for evaluating chewing detection algorithms.

Purpose of the Study:

• To introduce a novel multi-modal dataset for chewing activity monitoring.
• To facilitate the evaluation and advancement of chewing detection algorithms.
• To support research in understanding eating behavior through wearable technology.

Main Methods:

• Collected over 60 hours of multi-modal data from 14 participants in semi-free living conditions.
• Utilized photoplethysmography (PPG) and 3D accelerometer sensors, along with audio features.
• Provided detailed annotations and ground truth at both eating event and individual chew levels.

Main Results:

• A comprehensive dataset comprising raw sensor signals, extracted audio features, and precise annotations.
• Established a baseline evaluation method for chewing detection algorithms.
• Introduced a challenge to encourage improvements in existing chewing detection algorithms.

Conclusions:

• The presented dataset addresses the critical need for publicly available resources in chewing behavior research.
• This dataset and baseline evaluation will accelerate the development of more accurate chewing detection systems.
• Facilitates further research into dietary monitoring and health interventions using wearable sensors.