Cameras that catch violations: how it works and what they see
Author: DuoVision team
For: heads of HACCP, production, warehouse, and fuel station managers, and safety teams who want cameras to point out violations on their own
Summary
- A camera that catches violations is an ordinary IP camera plus a rule that describes the violation. When the rule triggers, the system creates a card with a frame, a responsible person decides whether it is a real violation, and the decision goes into a log.
- DuoVision, a Ukrainian video analytics company, does this on cameras you already have. No new cameras or new video surveillance system are needed, and frames are processed on a server at your site.
- In production it means checking hair covers, masks, gloves, crates on the floor, and riding on pallet jacks. In warehouses and fuel stations we set rules up per site and start in shadow mode; we recognize vehicle plates. We say in advance what a camera cannot see.
- To avoid drowning in false cards, every rule starts in shadow mode: cards arrive, nobody is punished for them, and after a few weeks you can see how far the rule can be trusted.
For: heads of HACCP, production, warehouse, and fuel station managers, and safety teams who want cameras to point out violations on their own
How an ordinary camera starts catching violations
The camera itself catches nothing. It films and streams. A program that reads the stream catches the violation. So you do not need new cameras to start: an IP camera that streams over RTSP or ONVIF is enough. See cameras for AI video surveillance for camera requirements. ONVIF Profile S, for instance, describes how video moves from an IP camera to a video management system.
The path from frame to result:
- Rule. Together with us you describe what counts as a violation at your site, for example "a crate stands directly on the floor" or "a person is in the zone without a hair cover".
- Card. When the rule triggers, the system does not sound an alarm across the plant. It creates a card: frame, camera, time, which rule.
- Specialist decision. A responsible person, such as the head of HACCP, looks at the card and decides whether it is a violation. The system only shows; it does not punish.
- Log. The decision goes into a log that shows where, how often, and on which shift violations happen. See the HACCP log sample for how it relates to the paper log.
The decision always stays with a person, because only a specialist knows what is normal at your plant.
Which violations cameras catch
In production
At the meat processing plant where we ran the system, rules checked the hair cover or net, mask, beard cover, gloves, a crate on the floor, and a person riding a pallet jack. Those are the rules we tested in practice. More: PPE control by cameras, crates on the floor, riding a pallet jack.
In warehouses and at docks
For warehouses and docks we set rules up per site: forklifts moving near pedestrians and truck dwell time. We start in shadow mode and promise no result in advance, because it depends on the camera angle. See the articles on the loading dock and forklifts and pedestrians.
At fuel stations
Rules for fuel stations are also set up per site. What can be monitored is described in video analytics for fuel stations.
Vehicle plates
We recognize vehicle plates at the gate: the system records plate, time, and a photo. It needs a camera aimed at the traffic lane. Details: plates at warehouse and elevator gates.
What cameras do not catch
- What is not in the frame. If shelving hides a person or a trolley hides the floor, the rule cannot answer. The card is marked "undetermined" instead of guessed.
- Faces. We do not recognize who someone is or search for people by face. The rule checks whether a hair cover is present, not whose head it is. On personal data: video analytics and personal data.
- Sound and weight. A camera does not hear speech or weigh cargo, so we do not promise that.
- Small details far from the camera. If a person is a few pixels tall, a beard cover cannot be checked. You need a closer camera or a different rule.
- Intent and reasons. The system sees a crate on the floor but not why. That is what the card and the person are for.
No rule should be called ready after a few good frames. Accuracy depends on the camera, light, and people, so it is checked at your site.
How to avoid drowning in false cards
This is the main fear of people who tried analytics before: the system produces hundreds of alerts and nobody reads them. What helps:
Shadow mode
A rule first runs "in the shade". Cards go to the responsible person, but nobody is punished or stopped. The person marks each one right or false. After a few weeks you see how many cards were wrong and why. Only then is the rule switched on, and if it fails it is retuned or dropped.
One zone instead of the whole site
Start with one zone and one or two rules. False cards are easier to sort out when there are few and they come from one camera.
A precise definition of the violation
Many false cards come from a vague definition. Our crate example: a crate on a pallet, trolley, or pallet jack is normal, and only a crate directly on the floor is a violation. So we agree on the line of a violation before launch and describe it for each camera. Another example: we first thought standing on the platform of an electric pallet jack was allowed, and the head of HACCP decided after viewing the frames that it is a violation. Agree on such lines before launch, not after.
Ask a vendor on how many separate events its accuracy was counted. A figure without a sample size means nothing. We do not promise absolute accuracy and quote no percentages until a rule is checked on new days. More: video monitoring of hygiene in a meat plant.
Why you do not need a new video surveillance system
Our approach is simple. We work on cameras that are already installed, so you do not buy new cameras or a new video surveillance system. Your cameras and recorders stay where they are, and the analytics read video from them. You connect only the rules you need. Frames are processed on a server at your site.
For what the cost is made of, see what video analytics costs.
Frequently asked questions
Does the camera punish employees by itself?
No. The system creates a card with a frame. A responsible person decides and owns any disciplinary step.
How long does shadow mode take?
Usually a few weeks for one zone. The exact time depends on how often events occur there.
Do I need the internet?
Frames are processed on a server at your site, so a constant internet connection is not needed for recognition. What remote access to the dashboard requires, we clarify at the site survey.
Can I start with one camera?
Yes. We run a pilot on one zone, and how many cameras it has depends on what each one sees.
Does it work at fuel stations and warehouses?
We set up rules for warehouses and fuel stations per site and start in shadow mode. It was proven in practice only at a meat processing plant, so for other sites the pilot shows the result.
Where to start
- Pick one zone and the violation that costs you most: hygiene, crates, vehicles, or plates.
- Write down what counts as a violation and what counts as normal at your site.
- Send us frames from that zone through the contact page.
- Run a pilot in shadow mode and judge the cards on your own frames.
It also helps to read how to choose video analytics: 10 questions to ask.
Where to apply this
- Food safety and hygiene
- Meat processing
- Occupational safety
- Warehousing and logistics
- Dish pass time in a restaurant: what to count and how a camera does it
- Restaurant kitchen control with a camera: sanitation, HACCP and what the camera checks
- Video analytics for a restaurant: what it does in the kitchen, at the pass and in the dining room