We present a testbed for exploring novel smart refrigerator interactions, andidentify three key adoption-limiting interaction shortcomings ofstate-of-the-art smart fridges: lack of 1) user experience focus, 2)low-intrusion object recognition and 2) automatic item position detection. Ourtestbed system addresses these limitations by a combination of sensors,software filters, architectural components and a RESTful API to trackinteraction events in real-time, and retrieve current state and historical datato learn patterns and recommend user actions. We evaluate the accuracy andoverhead of our system in a realistic interaction flow. The accuracy wasmeasured to 83-88% and the overhead compared to a representativestate-of-the-art barcode scanner improved by 27%. We also showcase twoapplications built on top of our testbed, one for finding expired items andingredients of dishes; and one to monitor your health. The pattern that theseapplications have in common is that they cast the interaction as anitem-recommendation problem triggered when the user takes something out. Ourtestbed could help reveal further user-experience centric interaction patternsand new classes of applications for smart fridges that inherently, by relyingon our testbed primitives, mitigate the issues with existing approaches.
展开▼