Vision AI's Decisive Moments: Offside Calls and Shelf Intelligence

If you've watched an offside decision come back within seconds at this summer's World Cup, you've seen vision AI at work — and it operates on strikingly similar principles to the technology watching shelves in retail stores. Here is Scandit's side-by-side comparison of the two systems.

Vision AI on the Pitch: Semi-Automated Offside

FIFA deployed a vision AI system to support offside decision-making. The setup:

• All 1,248 tournament players stepped into a scanning booth to create individual 3D avatars with precise limb dimensions.
• 16 stadium cameras capture player positions 50 times per second, tracking 29 body points per player.
• A sensor inside the ball transmits 500 times per second, pinpointing the exact moment the ball leaves a player's boot.
• Body positions at the moment of the kick are checked against the offside line, and decisions come back within seconds.

Vision AI in the Store: Shelf Intelligence

The same principles, aimed at a different goal: the perfect shelf. Correct stock levels, correct placement, correct pricing — with the planogram as the reference.

Just as football needed individual player scans because generic human shapes weren't precise enough, shelves need more than generic bounding-box detection. A bounding box can tell you something large is present; it cannot tell you whether a specific SKU is correctly placed and properly oriented. That takes SKU-level models trained on product design, dimensions, and orientation.

Fixed cameras are joined by associates' mobile cameras and autonomous robots to track shelf state. When errors are detected, prioritized alerts guide store associates to restore products to their correct positions — before customers notice.

The Two Systems, Side by Side

Technology dimension	Offside monitoring	Shelf intelligence
Camera setup	Fixed stadium roof cameras covering the pitch	Fixed, mobile, and robot cameras covering the store
Object detection	Each player tracked as a unique object	Each SKU tracked as a unique object
Tracking model	29 skeletal keypoints per player	SKU-level models trained on design, dimensions, orientation
Reference comparison	Player position vs. offside line	Shelf state vs. planogram
State change detection	The millisecond the ball leaves the boot	Out-of-stock, misplaced, or mispriced products
Automated alerting	Offside alert to VAR officials	Prioritized alerts to store staff
Decisive action	Officials make the call	Associates complete guided fixes

Both systems work because of fixed geometry. The pitch stays constant and cameras capture consistent images, so every measured position references a stable frame. Shelf intelligence operates identically: cameras photograph the same view, shelves stay put, and planograms define what "correct" looks like. Everything vision AI sees is seen in context — deviations from the reference state become a simple matter of math.

The Key Difference: One Moment vs. Continuous Decay

One significant distinction remains. Football captures a single moment — the ball's departure triggers the call, and speed drives everything.

Shelves pose a harder problem. They degrade continuously through operating hours as products sell and errors accumulate. No single moment defines the problem; the cost builds quietly over time.

Yet both systems solve the same prioritization challenge. Officials don't need every player's position every second — they need an alert when someone crosses a line. Associates don't need a notification for every shelf image or minor misplacement — they need to know when a best-seller drops below threshold during peak hours, or a high-velocity SKU disappears from the shelf. The right call at the right moment, without the noise — the goal is identical.

Toward the Perfect Shelf

This summer's tournament shows what fixed cameras, trained object models, and reference states make possible: previously impossible decisions at the required speed and accuracy.

Retail has understood the principle for years. The shelf is the pitch. The planogram is the offside line. Products are the players. And the camera keeps asking the same question: Is it where it should be?

For Korean stores, the practical starting point needs no ceiling cameras at all — associates' smartphones can scan shelves today. For the technical background, see computer vision AI in retail; for store scenarios, see our retail industry page. Data Connect's technical team supports adoption reviews.