Computer Vision в ретейле: «Лента» распознаёт ценники с движущегося робота

Retail chain Lenta (the LentaTech team) turned the task of recognizing price tags from a moving robot's camera into a Computer Vision ML hackathon in 2026 — aiming to automate shelf monitoring and catch violations right on the sales floor.

Why is this harder than a static camera?

Lenta mastered recognition from a static camera long ago and already uses it in its stores. A moving robot's camera is a fundamentally different task: the frame blurs, the angle and distance to the shelf constantly change, and lighting in the store is uneven. Because of this, reading price tags accurately "on the move" turned out to be noticeably harder to implement than the familiar recognition from a fixed point. This was the hard part the team had to solve before the scenario could work in a real store.

Quality recognition of price tags from a moving robot's camera sounds

much more interesting and challenging, especially to implement.

> — from the LentaTech team's corporate blog on Habr

What the task technically involves

Recognizing a price tag from a moving platform is a chain of several Computer Vision subtasks. First, the model detects the price tag itself in the frame, then reads the text — the name, the price, the barcode. Separately, it needs robustness to motion blur and glare, plus linking the recognized price tag to a specific spot on the shelf to understand which product it belongs to. On a static camera, some of these steps are simplified by a fixed angle; on the robot, every frame arrives from a new angle, so the model has to be more robust.

Which processes the scanning robot covers

A robot with a camera covers several store processes at once. Staff spend less time manually walking the aisles and monitoring shelves, while the robot catches violations on the move — for example, missing or incorrect price tags — and passes this information to the team. According to LentaTech's description, this takes the routine off staff and speeds up the response to problems with merchandising and pricing.

Key points from the publication:

• The task — recognizing price tags from a moving robot's camera on the sales floor
• The base — experience with static-camera recognition that Lenta has long applied in its stores
• The goal — reduce staff time spent monitoring shelves and automatically flag violations
• The format — a Computer Vision ML hackathon on data captured by the robot

What this means

Computer Vision in retail is shifting from static cameras to mobile robots: recognition "on the move" turns shelf walks from manual routine into an automatic process. Lenta's case shows that even a well-honed CV task has to be reinvented once the camera starts moving — and that businesses are willing to hand such tasks to hackathons to assemble a working solution faster.