What does image recognition actually detect on a shelf?
On a shelf, image recognition detects the products themselves, their stock levels, how they are placed, and shopper interaction with them. Computer vision is the underlying technique: computers use repeated exposure to digital imagery to identify items in real time from a live feed 1.
What gets detected
- Product identity, with the vision model recognising individual items from a camera feed as they appear on the shelf 1.
- Stock depletion, where AI trained algorithms use shelf pictures to alert the stock room once a product reaches a pre-determined level, so the item stays available 2.
- Product placement and visibility, evaluated by AI-driven algorithms in mobile applications that analyse photographs of store shelves 3.
- Shopper engagement with the shelf, including the specific product a shopper has picked up 4.
How the detection is built
Accuracy comes from training volume rather than from any single camera. It takes about 1 million images of the same item for a computer vision product to reach 99% accuracy in identifying products in real time 1. Some providers supplement cameras with smart shelving and RFID tagging to identify when products are removed from shelves, though that approach is specific to the vendor while AI enabled cameras are constant across providers 1.
Where shelf detection can fail
The shelf itself is part of the detection problem. In one scenario, shelves made of a highly reflective material caused a robot's vision software to see something that tricked it 5. Related image-based detection exists outside the shelf context too: a photo-based AI tool called Vision examines product images in minute detail to predict counterfeit likelihood 6.