Computer Vision for Loss Prevention vs Shelf Monitoring: Different Systems

One camera, two very different jobs

It is tempting to assume that once a store has cameras and computer vision for shelf monitoring, loss prevention is a small add-on. It is not. Loss prevention and shelf monitoring are different systems that happen to share a sensor type, and treating one as a feature of the other leads to a deployment that does neither well and carries risks it did not budget for.

What each system looks at

Shelf monitoring looks at products — is the facing full, is the right item in the right place. Loss prevention looks at people and behaviour — concealment, exit without payment, sweep events. The models are trained on different things, and a shelf model has no useful signal about theft.

Camera placement is different

Shelf monitoring wants a stable, consistent view of a fixture, usually from the aisle. Loss prevention wants coverage of entrances, exits, high-risk fixtures, and blind spots, often from above and at angles that would be useless for reading a shelf. You cannot serve both from one camera plan.

Data handling diverges sharply

Shelf monitoring, done well, processes on-device and sends a structured event with no personal data — the approach in our computer vision project. Loss prevention necessarily involves people in frame, may retain footage as evidence, and can feed decisions that affect individuals. That pulls in a much heavier set of privacy, retention, and legal obligations, and often staff and works-council considerations.

Different stakeholders, different accountability

Shelf monitoring answers to store operations and supply chain. Loss prevention answers to security and legal, with policies about how footage is used, who can review it, and how an intervention is authorised. Merging the systems muddies that accountability.

Why the coupling is tempting and still wrong

Shared hardware cost is the draw. But the incremental cost of separate camera plans and pipelines is small next to the cost of a loss-prevention capability that was scoped as a shelf-monitoring line item and never got the legal review it needed.

A worked example

A retailer with shelf-monitoring vision in 200 stores is asked to "just add theft detection." Doing it properly means a separate camera survey for entrances and high-risk bays, a different model, a footage-retention policy, a privacy assessment covering people in frame, and sign-off from legal and security. It is a distinct project with its own budget and timeline — which is the right outcome, because bolting it onto the shelf system would have shipped an unreviewed surveillance capability.

Shared infrastructure is the only safe overlap

The one thing the two systems can genuinely share is physical: mounting rails, power runs, and network cabling, planned at install so that a future loss-prevention project is not re-cabling the store from scratch. Everything above that layer stays separate — cameras chosen for the job, models trained for the job, pipelines, storage, retention rules, and access control. Deciding the shared physical layer up front is cheap foresight; letting it blur into shared cameras and shared footage is how a stock tool quietly becomes a surveillance system without the review that requires.

Where this stops being right

• A store that genuinely wants both should run them as two projects with shared mounting infrastructure at most, not one system.
• Small operators may buy a packaged loss-prevention product rather than build; that is a separate procurement from shelf monitoring.
• Any people-facing analytics — dwell time, demographics, staff monitoring — is a third category again, with its own assessment.

FAQ

If a store already has shelf-monitoring vision, is loss prevention a quick add-on? No. It needs different cameras and placement, a different model, footage retention and privacy handling, and legal and security sign-off. It is a separate project.

Why not run both from the same cameras? Because shelf monitoring wants a stable aisle view of a fixture and loss prevention wants entrance, exit, and high-risk coverage from other angles. One camera plan cannot do both.

What is the real risk of merging them? Shipping a surveillance capability that affects people without the privacy assessment, retention policy, and legal review it requires, because it was scoped as a stock-monitoring feature.

ISTRALLEN builds shelf-monitoring computer vision as its own system and keeps loss prevention a separate, properly scoped project — see AI for Retail.