Automatic dietary monitoring
Nutrition is fundamental to health and paces everyday routines. Dietary monitoring is needed in many healthcare areas, from diabetes and dementia to overweight and obesity. Our research aims to design effective and practical sensor devices and signal pattern analysis algorithms that explain diet-related behaviour. Whereas current dietary monitoring requires respondents to manually enter data, whether in paper forms or smartphone apps, our intention is to infer as much dietary data as possible from sensor measurements, hence automatic dietary monitoring. We expect that less manual labour and unobtrusive monitoring devices will yield better long-term wearing compliance and more accurate behaviour information for coaching applications.
The first demonstrations of wearable sensors for food and drink intake motion monitoring, chewing sound interpretation, and swallowing detection were made in our publications during 2005-2006. More detail on the sensors and signal interpretation can be found in the respective subsections below.
Subsequently to the first confirmation that wearable sensors could indeed reveal diet-related behaviour, our work focused on structuring dietary information on intake timing and schedule, food type and amount, calories and nutrients, and micro-structure analyses. We are continuing to investigate automated dietary monitoring and coaching in several projects.
"Automatic Dietary Monitoring Using Wearable Accessories", Seamless Healthcare Monitoring: Advancements in Wearable, Attachable, and Invisible Devices, Springer, 2018.,
"Ambient, on-body, and implantable monitoring technologies to assess dietary behaviour", International Handbook of Behavior, Food and Nutrition, Springer, 2011.,
"On-body sensing solutions for automatic dietary monitoring", Pervasive Computing, IEEE, 2009.,
During intake, bites, chews, swallows, and movements occur in an orchestrated structure. We investigated different analysis and modelling techniques to describe the phenomena and track the events. The event structure of movement, chewing, and swallowing was modelled using probabilistic context-free grammar models (PCFG), which could deal with recursive structure found in the chewing and swallowing events.
To understand the changing material structure and acoustic patterns during chewing, temporal sequences were searched using multi-objective evolutionary search strategies. Results indicated that the temporal structure can be found in most raw and brittle foods. Often a two-phase structure was observed. However, soft-texture foods exhibited a one-phase structure only.
"Probabilistic parsing of dietary activity events", BSN 2007: Proceedings of the International Workshop on Wearable and Implantable Body Sensor Networks, Springer, March 2007.,
"Automatic Identification of Temporal Sequences in Chewing Sounds", BIBM 2007: Proceedings of the IEEE International Conference on Bioinformatics and Biomedicine, IEEE Press, November 2007.,