SmrtFridge: IoT‑based, User Interaction‑Driven Food Item & Quantity Sensing
Links:
Problem: Modern consumers struggle to keep track of fridge contents without tagging food items. Existing smart fridge solutions rely on RFID or manual labeling, making adoption burdensome.
Solution & Contributions: SmrtFridge is a consumer‑grade prototype that identifies what users place into or remove from their fridge, and estimates how much remains in a container, all without RFID tags or explicit user input. It uses a hybrid sensing pipeline:
- Item Segmentation: Proposes two methods including
- Optical-flow based item segmentation (uses only RGB images/video)
- Thermal image based segmentation (uses thermal + RGB)
- Item Classifiation: A Deep Neural Network that classifies the extracted food images.
- Quantify Estimation: Thermal contrast measurement (via temperature differences) to infer coarse levels of remaining food (e.g., empty/half/full).
System Architecture:
Results: In a user study with 12 participants and 19 common food items, SmrtFridge correctly captured at least 75 % of a food item’s image in over 97 % of interactions. It achieved approximately 85 % precision and recall in identifying food items, and around 75 % accuracy in three-level quantity estimation.
Impact: SmrtFridge demonstrates that ambient sensing techniques can provide accurate food recognition and quantity tracking without intrusive tags—paving the way for more accessible smart kitchen applications.