Real-time Bite Detection from Smartwatch Orientation Sensor Data
Christos Maramis, Vassilis Kilintzis, Nicos Maglaveras · 2016
The introduction of smartwatches over the last few years has made widely available a new type of wearable, everyday-usage device that is equipped with dozens of sensors. The sensors embedded in the smartwatch constitute a valuable source of data about the bodily functions of the smartwatch user; a source that has already been exploited for inferring information concerning the human behavior. One possible application of the aforementioned information inference is the prediction of eating-related events, such as bite instances. Accurate bite instance prediction from smartwatch sensors could serve as a trigger for appropriate user feedback in the context of just-in-time adaptive interventions for eating behavior management. In this paper, we present a novel method for real-time detection of bite instances from 3-axis orientation data acquired by a smartwatch. The evaluation of proposed method has been performed on eight annotated orientation timeseries, generated by eight individuals who wore a commercial smartwatch on their active hand while eating a bowl of milk with cereals. Both the classification accuracy of the method and its ability to make real-time decisions were evaluated, yielding very promising results.