(19)Fresh off Patent
(45) Week of Oct 5, 2026 No. 5 3,852 lapsed 197 picks
Software & AI · lapsed Aug 28, 2026

Point-and-tap camera recognition for plated food items in a cafeteria

A camera-based checkout aid that shows a live image of a tray of multiple food items, lets an operator tap a location to place a fixed-size selection frame around the assumed item, runs feature-based recognition inside that frame and returns the top candidate, and if the operator requests a change it offers alternate candidates. It is aimed at cafeterias or canteens where staff or users quickly select items from a tray rather than rely on container recognition.

Sheet 1 of the patent drawingsSheet 2 of the patent drawingsSheet 3 of the patent drawings

(57) What the patent covered

The device captures an image of multiple commodities, displays it, counts whether a selection frame already exists, when the operator taps it either creates a fixed-size frame around the tapped position (if none) or interprets a subsequent input as a position check, runs feature-based recognition on the image area inside that frame to produce ranked candidate commodities, outputs the top-ranked candidate's information, and if a change instruction is received, outputs an alternate candidate selected from the ranked list.

The distinctive part: Tightly couples a simple tap-driven fixed-size frame UI with on-frame feature matching and a ranked-candidate change flow intended for fast cafeteria checkout.

Novelty 3/5. When filed (around 2013) computer-vision recognition and candidate ranking were known, but integrating a tap-to-place fixed-size frame and a simple change-instruction to pick lower-ranked candidates for cafeteria tray workflows was a practical UI-plus-recognition combination rather than a wholly new algorithm.

Building it

A small team would mount an off-the-shelf USB camera above the tray area, build a web or native UI that streams the camera and implements a tap/click that draws a fixed-size rectangle, run a lightweight CNN or classical feature extractor (SIFT/ORB or MobileNet features) on the pixels inside the rectangle to produce embeddings, compare against a local catalog of item embeddings to get ranked candidates, and show the top match with an "other" button that cycles alternate candidates. Use a small SQLite catalog and a simple REST or local inference server for the model.

Components: RGB camera (mounted over checkout area), Display with touch or click input, On-device processor and memory, Recognition model/dictionary and feature-extraction software

Buildability 5/5. Straightforward hardware and software; the main work is collecting representative images per menu item and tuning the recognition model for varied plating and occlusion.

Is anyone buying?

Comparable products exist; none doing exactly this was found. Market has several documented entrants focusing on vision-based tray/dish recognition for cafeteria self-checkout; some vendors emphasize AI-driven recognition for trays, with implementations in campus, corporate, and large-scale cafeteria settings.

This demand check was run with a web search in the week of Oct 5, 2026. Listings change often, so treat it as a starting point for your own research.

Who buys it: cafeterias, school dining services, corporate canteens

Before you make it

We didn't find any related continuation or divisional patents still in force, although that doesn't rule out design patents, trademarks or unrelated patents covering the same product.

Accuracy depends on good training images and consistent camera setup; crowded market for food-recognition models and potential privacy or food-labeling regulations to consider.

Owners occasionally revive lapsed patents (about 1.5% are), and this page is a research lead rather than legal advice, so get a freedom-to-operate opinion before you manufacture anything.

Claim 1, as granted

1. A commodity recognition apparatus, comprising: a memory configured to store computer executable instructions; and a processor configured to execute the computer executable instruction to perform operations, the operations comprising: photographing an image capturing area including a plurality of commodities to capture an image where the plurality of commodities are imaged; displaying the image where the plurality of commodities are imaged on a display screen; receiving a selection input on any position on the image; counting a number of frames displayed on the image; displaying a frame of a predetermined size on the image to distinguish, from the other commodities, one commodity which is assumed to be displayed on the position on which the selection input is received as a recognition object in response to the selection input when a count value obtained by the counting is zero; determining a position on which an instruction input is received in response to the selection input when the count value is one or more; displaying a frame of the predetermined size on the image to distinguish, from the other commodities, one commodity which is assumed to be displayed on the position on which the selection input is received, as a recognition object if the position on which the instruction input is received is outside the frame; recognizing a candidate of the commodity assumed to be displayed in the frame according to the feature amount of the image in the area surrounded by the frame; outputting information of a best candidate commodity among the candidates of the commodity recognized; receiving a change instruction for the best candidate commodity for one commodity assumed to be displayed in the frame if the position on which the instruction input is received is inside the frame; and outputting information of the commodity other than the best candidate selected from the candidates if the change instruction for the candidate of the commodity is received.