Retail computer vision guide
Retail computer vision and shelf scanning explained
Shelf scanning converts store images into structured observations: where products appear, how many facings are visible, which packages may match, and where human review is still required.
Short answer
Retail computer vision can detect and crop products quickly. Exact SKU recognition is harder because flavors, formulas, and sizes often share nearly identical packaging. Reliable shelf intelligence combines visual detection, identifiers, package text, product sources, and review.
The main shelf-scanning tasks
- Product detection: draw a box around each visible product facing.
- Product grouping: suggest which facings appear to be the same exact item.
- Product recognition: connect a crop to a brand, product, variant, and size.
- Shelf intelligence: measure availability, facings, placement, and possible gaps.
- Catalog creation: turn accepted matches into store-ready product records.
Why grocery shelves are difficult
Retail shelves contain reflections, price tags, partial packages, repeated facings, changing artwork, small text, and near-identical variants. A detector may correctly find a package while a recognition system still confuses two flavors. Geometry and visual similarity are not enough to guarantee exact identity.
What a trustworthy system records
A good workflow preserves the original image, normalized crop coordinates, detected identifiers, visible text, external product evidence, confidence, and reviewer decisions. It separates the observation—“a package was visible here”—from the trade-item identity ultimately assigned to it.
Where Nearby applies it
Nearby uses retail computer vision as part of store digitization: capture products from shelf photos, prepare source-backed catalog candidates, and let owners approve what becomes sellable. Explore Nearby Store Intelligence or get my store online.