Glove checks in meat processing: what a camera verifies and where it sees poorly
Author: DuoVision team
For: heads of HACCP, process engineers, and supervisors of deboning, trimming and packing at meat processing plants
Summary
- The camera checks for a glove on a hand that is visible in the frame. If the hand cannot be seen, the correct answer here is "undetermined".
- Gloves are either disposable (for product hygiene) or chain mail (for protection from a knife cut). The system we tuned checked that a glove was present. It did not judge protection class or condition.
- Order No. 590 requires staff conduct rules and clean work clothing, but it does not name gloves. The plant writes down when and which gloves are worn in its own programs.
- In our measurement of 219 cut-outs (131 events, 8 cameras, one day), about 23% of "glove present" answers were not supported by the frame (between reviewers, from 8 to 56%). A camera that films people from behind sees no hands at all.
- We have no independent test on new days, so we do not quote an accuracy figure for gloves.
For: heads of HACCP, process engineers, and supervisors of deboning, trimming and packing at meat processing plants
Why check gloves in deboning, trimming and packing?
There are two reasons. The first is about the product. Hands touch raw meat, knives, tables and crates. If they are not washed and the gloves are not changed, they carry microorganisms to the next piece. Our article on listeria in a meat plant cites data that listeria was found on the hands of food workers far more often than on office workers. How rarely hands get washed without supervision is covered in our piece on handwashing in production.
The second reason is about people. According to US injury statistics (BLS, 2015 data, cases with days away from work), hand injuries made up 25% of such cases in meat processed from carcasses, and cuts and lacerations made up 14% of cases among slaughterers and meat packers. These figures do not separate the causes, but they show where injuries hurt most often in meat production.
How do disposable gloves differ from chain mail gloves?
They are two different items with different purposes.
| Type | Purpose | What a camera can check |
|---|---|---|
| Disposable (nitrile, vinyl, polyethylene) | Product hygiene: a barrier between hand and meat | Whether a glove is on the hand |
| Chain mail (metal mesh) | Protecting the hand that holds the meat from a knife cut | Presence on the hand, if the hand is visible. Protection class and condition of the mesh cannot be seen |
For chain mail gloves and arm guards there is the European standard EN 1082-1 (protection against cuts and stabs by hand knives), and for aprons, trousers and vests there is EN ISO 13998. Both are occupational safety standards, meaning requirements for the product. They do not say whether a worker is actually wearing it. We have no measurements of our own on chain mail gloves: our rule checked gloves as work clothing.
What do the regulations require about gloves?
Order No. 590 of the Ministry of Agrarian Policy and Food (October 1, 2012), clause 2.10.3, requires that work clothing does not contaminate the product and that procedures exist for laundering and verifying it. Clause 2.10.5 lists staff conduct rules, including handwashing and what to do after cuts. In the clauses we read, the word "gloves" does not appear, so the plant decides in its own programs which gloves to wear and when to change them.
Regulation (EC) No 852/2004 (Annex II, Chapter VIII) requires everyone working in a food-handling area to keep a high degree of personal cleanliness and to wear suitable, clean and, where necessary, protective clothing. Gloves are not named there either.
The practical consequence is that what counts as a violation depends on your records and rules. We configure the system to your rules and do not impose ours.
How does a camera check that gloves are worn?
The system finds a person in a zone where gloves are mandatory, then looks for their hands and checks whether there is a glove on each. If a hand is visible and there is no glove, an event card with the frame is created. The decision is made by the responsible person, for example the head of HACCP. The general scheme is described in our overview of PPE control by cameras.
The main rule: we establish a missing glove only when the hand is really visible. A hand hidden by a table, by product or by the person's own body is never counted as bare. That answer is "undetermined". Without this, the rule would raise cards against people who are wearing gloves and simply stand with their back to the camera.
Why are gloves hard to see from angled cameras?
A glove is small. In a wide frame from under the ceiling, a hand takes up a few dozen pixels, and it is covered by the table, the knife, the product and the other hand.
The angle matters even more. One camera filmed workers from behind and above. On the camera that filmed backs, the system answered "not sure" in 55 of 68 cases (cut of 21.09). We looked at six of them. In five, the person had their back to the camera, their hands were in front of the body, and no hands were in the frame. Another camera in the same plant, filming people from the front, gave only 16% of such answers, and blue gloves were clearly visible on it. That camera never produced a single card about gloves or masks in its whole history.
For other rules this camera produced cards. For gloves the best camera is one that films hands from the front and from close up.
What did we measure, and what turned out to be wrong?
At the meat processing plant where we ran the system, we suspected that two deboning cameras were worse than the rest: they seemed to produce more "glove present" answers with no basis. To check, we took all 219 cut-outs with a "glove present" answer from eight cameras (131 independent events, one day, a shift of 3 hours 14 minutes). Nine independent reviewers, 25 frames each, reviewed them blind: without the camera name and without knowing the hypothesis.
The result: on 169 cut-outs (77%) a hand was visible and the glove was there. On 24 (11%) a hand was visible but there was no glove, or it could not be made out. On 26 (12%) there were no hands in the frame at all. So about 23% of answers were not supported by the frame.
The hypothesis about the two bad cameras was not confirmed: 13 of 56 cut-outs (23%) on the suspect cameras, 37 of 163 (23%) on the rest. Earlier, on 45 cut-outs, we had seen 47%, but those images were reconstructed afresh and differed from the ones the system actually looked at. The gap showed us that a conclusion from a small sample needs checking.
Limits of the measurement. All 219 cut-outs come from one day and one shift. The figure itself is shaky: each of the nine reviewers judged their own 25 frames, and the shares ranged from 8% to 56%. Our own check on 8 frames matched the reviewers in 5 cases out of 8, and the disagreements were on the edge between "glove present" and "cannot tell". Comparing cameras with each other is more reliable than the number itself. We have not run an independent test on new days.
One more figure from the history of the rule. All four glove cards that the responsible person reviewed were confirmed. Four cards is very little, and we do not call it accuracy.
What can a camera not tell you about gloves?
- Whether a chain mail glove matches the required protection class, or whether the mesh is torn. A disposable glove with a puncture also looks intact.
- Whether gloves were changed after a break or after touching the floor or raw product. The camera sees the glove on the hand at that moment and knows nothing about its past.
- What happens in places no camera sees: behind a column, under a table, in a blind corner.
- A glove on a hand that is not visible is not assessed.
We do not promise 100% detection. For gloves it is realistic to talk about cards where the hand is in the frame and about "undetermined" where it is not.
Where should you start at your plant?
- Write down where gloves are mandatory: deboning, trimming, packing, portioning. And where they are not needed, so you do not get pointless cards.
- For each camera, check whether workers' hands are visible from there. A simple test: pause the footage and see whether you can tell a glove apart without zooming in.
- Decide on the type: which glove the system should treat as normal and whether colour has to be distinguished.
- Send us a few hours of footage. We will show you on which cameras a glove can be read and on which the answer will be "undetermined".
The first stage runs in shadow mode: you see the cards and the workers do not. What this looked like in a real launch is described in our lessons from launching in a meat plant.
Where to apply this
- Food safety and hygiene
- Meat processing
- Food manufacturing
- 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
Sources
- Order of the Ministry of Agrarian Policy and Food of October 1, 2012, No. 590 «Про затвердження Вимог щодо розробки, впровадження та застосування постійно діючих процедур, заснованих на принципах НАССР» (“On Approval of the Requirements for the Development, Implementation, and Application of Permanent Procedures Based on HACCP Principles”), clauses 2.10.3 and 2.10.5
- Regulation (EC) No 852/2004 on the hygiene of foodstuffs, Annex II, Chapter VIII (Personal hygiene)
- U.S. Bureau of Labor Statistics, Monthly Labor Review, 2017: How safe are the workers who process our food? (2015 data)
- EN 1082-1:1996 Protective clothing. Gloves and arm guards protecting against cuts and stabs by hand knives. Part 1: Chain mail gloves and arm guards
- EN ISO 13998:2003 Protective clothing. Aprons, trousers and vests protecting against cuts and stabs by hand knives