food recognition

AI FOODYPOS

The project aims to create a system for automatic food recognition. Applying image processing techniques, an acquisition station captures - at the right moment - few still images of a tray with meals and drinks appearing in the field of view of an RGB camera above it. Subsequently, using deep learning paradigms, a specific module of the system locates and recognizes each element composing the meal, in a fine-grained fashion. The way people arrange the plates on the tray is very different, as well as the  number and type of items chosen, and the possible presence of non-food elements. Furthermore, the appearance of food is very different depending on the specific meal, the presence of topping or sauce, the nearness of different foods inside the same plate.  All this makes the problem very challenging and interesting.

The modules developed by TeV can be integrated, for example, in an automatic check-out system for corporate or student restaurants, but also as a checker for controlling the exact correspondence of the served meals with respect to those booked.

Duration: from June 2020 to May 2021 (1y)

Partner: Sidera ICTease