Validating OCOsense smart glasses in a three-week home-based study: Assessing detection of eating, food...

We investigated the potential of OCOsense™ smart glasses to monitor eating and support behaviour change in everyday life. Twenty-three volunteers (14 women, 7 men and 2 non-binary individuals) wore the glasses for 7 h/day, at least five days/week for three weeks. To establish a baseline, in week one, participants wore the glasses with minimal intervention; in week two, they annotated each eating event and photographed their food to evaluate detection accuracy. In week three, participants responded to real-time notifications confirming whether eating was occurring when sensed by the glasses, and received haptic feedback (i.e., glasses vibration) prompting them to either chew more slowly, chew more per mouthful or take longer pauses between bites. Based on sensed chewing, the glasses algorithm detected 476/498 eating events logged in week two (F1-score 0.89), and 528/548 real-time events in week three (F1-score 0.91). To assess food detection accuracy, 598 images (1036 items) were analysed by an algorithm and compared to manual inspection. The algorithm correctly identified 919/1036 items. Compared to baseline, those aiming to change their behaviour: 1) reduced chewing rate (1.63-1.57 chews/sec), while others increased slightly (1.70-1.73); 2) increased chews per bite (22.0-22.8), which declined in others (22.8-19.5); and 3) lengthened breaks between bite (4.69-5.33 s), while others shortened theirs (5.16-4.85 s). Overall, this study demonstrates that OCOsense glasses can detect when and what people eat in everyday life and that feedback delivered by the glasses has the potential to support behaviour change.