Computer vision: matching shelf stock to the system
Retail, store chain
Industry and type of organization
A large retail chain. The stock recorded in the system and the real stock on the shelf could drift apart, and the discrepancy usually surfaced too late.
Business problem
Between what the system shows and what really stands on the shelf, a gap opens up: goods misplaced, not scanned, moved elsewhere. With a single store you catch it on a walk-through. With a chain of stores, manually checking every shelf is unrealistic, so discrepancies grow until they turn into gaps on the shelf or into orders based on a wrong stock level.
Approach and scope of work
Using transfer learning we adapted a computer vision model that compares a shelf photo with the stock recorded in the system. The model combines detection and object recognition: it detects the individual products in the photo, their position and what they are, and then sets that picture against the system data.
The check no longer requires walking every shelf. The comparison happens in the background, and the employee goes straight to where the image and the system disagree, instead of checking everything in turn.
Business effect
photo → stock
automatic comparison of the shelf with the system
instead of a manual walk-through store by store
discrepancies
caught before they grow into gaps
checked first
The system points to the places where real and system stock differ, so the check goes straight to where the problem is. Discrepancies surface earlier, before they turn into an empty shelf or an order calculated on a wrong stock level.
Scale of the implementation
The comparison works at the level of a single shelf and covers the store chain. The model recognizes the image and sets it against the current stock in the system.
If the stock in your system does not always match what is actually on site, describe your case to us. We will check whether the image can be matched against the data automatically.
Related articles in the cm-opti knowledge base
Service area: Computer vision and image analysis.