Blog / MES
MES · Guide

Computer Vision for OEE: What Cameras Can Measure

OL By OEE Lab Editorial |Updated 15 November 2026 |6 min read
Short answer

A camera on the line can count products where a machine has no usable signal. Stops come from the PLC, sensors or the operator. Where cameras are fitted, video of the minute of a stop can be streamed on request next to its machine data, so the team sees what happened, not only how long it took. It adds to PLC data; it does not replace clean signals where you have them.

What cameras can measure

A camera pointed at a line can do a few jobs that matter for OEE. It is not magic.

It sees only what is inside its view, and it keeps counting when people are busy.

JobWhat the camera doesLimits
Count productsCounts items crossing a line or entering a zone, also where a machine has no usable signalNeeds a clear view and a test against a hand count
Time a stopNot the camera's job. Stop timing comes from the PLC, an IoT sensor or the operatorA PLC bit shows time but not reason
Show the causeVideo of the minute of a stop can be streamed on request, next to its machine dataNot stored, so someone has to watch it and tag the cause
Measure speedCounts items per minuteNeeds a stable frame rate and view

The value is in the first and third rows. A signal can say a line stopped for 38 seconds.

Watching the minute of that stop can show that a label curled.

Why the cause is the hard part

Short stops are the classic blind spot. They are too short to log by hand, and a PLC bit shows time but not reason. See micro-stops and the hidden factory.

Operators do clear them, but they cannot write down 30 of them a shift and keep the line running. The PLC logs each one. Where a camera is fitted, the team can stream the minute of a stop on request, tag the cause and group the stops by cause.

Example: from 18 stops to a fix list

Example: a packaging line logs 18 stops of about 40 seconds in a shift. The PLC data says only "stop".

The PLC logs each stop. The team streams the minute of the most frequent stops live, on request, and tags the cause. The daily meeting reviews the tagged causes.

Label misfeed at the same spot11 stops
Bottle falls on the conveyor4 stops
Other3 stops
The 18 stops, grouped by tagged cause. One cause is more than half of them.
Time lost: 18 stops x 40 s = 720 s = 12 min
Share of a 480 min shift: 12 / 480 = 2.5%
Label misfeeds alone: 11 x 40 s = 440 s = 7.3 min = 1.5%
Fix the label feed and you win back about 1.5 points of the shift, from one cause.

This is where OEE work gets practical. The 12 minutes were already counted against OEE as lost time.

Tagging turns them from a vague "stop" into one named problem. For the next step, see how to reduce micro-stops.

Camera, PLC or sensor?

If this is trueStart with
The machine has a PLC and you can read itPLC signals, then add a camera to see what happened
The machine is old with no outputA sensor retrofit, or a camera if the flow is visible
Stops have unclear causes that repeatA camera on the problem spot
The product leaves the machine in a clear streamA camera can count it
The area is dark, steamy or very dustySensors first, and test a camera before relying on it

The best setups mix them. The PLC gives exact timing and fault codes.

A camera gives the picture. Read more in PLC signals for OEE and retrofit sensors for OEE.

How to pilot it on one line

1
Pick the spot
Choose the point where most stops happen. Mount the camera so it sees the product, not the people.
2
Log for a week
Let the PLC or a sensor log stop times while the camera counts. Do not change anything yet.
3
Check the counts
Compare the camera count with a hand count for one hour. Set your own tolerance.
4
Tag the stops
Stream the minute of frequent stops on request and tag the cause. Agree names with operators.
5
Fix the top cause
Make one change, then compare stops per shift before and after.
A five step camera pilot on one line.

What to do with what you see

Watching a stop has no value until it changes a decision. A simple weekly routine works well:

  • Monday: list the top three tagged stop causes from last week.
  • Daily, five minutes: the shift lead walks the crew through the worst tagged stop of the day.
  • When a fix ships: compare stops per shift before and after, on the same product.
  • Monthly: retire causes that are gone, and name any new ones.

Keep the list short. Three causes at a time is enough.

The goal is fewer stops, not more watching. For the daily meeting format, see daily production meeting.

Privacy and trust

Tell people before you install a camera. Aim it at the product and the machine, not at faces.

Agree who may watch the streamed video. Check local law and works council rules where they apply.

The goal is the process, not the person. A camera that is used to blame will be covered with tape within a week.

Common mistakes

  • No test against a hand count. Trust the count only after you have checked it.
  • Poor light. Glare and shadows change what the camera sees. Fix the light before the model.
  • Replacing good signals. Where a PLC already gives a clean count, use it. Use the camera for what it adds.
  • Watching stops without tagging a cause. Tag each stop you watch, review the tagged causes in a daily meeting, then act.
  • Moving the camera without a re-test. A bump or a new bracket changes the view. Re-check the count after any move.
  • Training on one product only. A new pack size or color can change what the model sees. Test each new product before you rely on it.

For OEE basics, see what is OEE, and for fault-finding on camera systems, vision inspection troubleshooting.

See it live in Fabrico

Where cameras are fitted, Fabrico shows video of a stop, streamed on request next to its machine data, so short stops become a fix list. See how Fabrico uses cameras for OEE.

What people ask next

Can a camera calculate OEE by itself?

Where it has a clear view of the product flow, it can supply the output count, a core input for performance. Stop times come from the PLC, a sensor or the operator. OEE also needs the ideal rate and a quality count. See how to calculate OEE.

Do I need a camera if I already read the PLC?

Not for timing. Where a camera is fitted, streaming the minute of a stop on request adds a view of what happened, which a PLC bit cannot show, mostly for short stops that repeat. Many plants use both.

How do I check that a camera count is right?

Compare it with a hand count for an hour, at normal speed, and repeat it at different times of day. Set your own tolerance before you start.

Is it allowed to put a camera on the line?

Rules differ by country and site. Tell staff, point the camera at the product, agree who may watch the streamed video, and ask legal or the works council where needed.

What kind of stops are worth watching on camera?

Short stops with a visible cause, such as jams, misfeeds, falls and blocked chutes. See how to reduce micro-stops.

← More OEE Lab guides

OEE essentials: OEE calculator · What is OEE · Six big losses · Availability · TEEP · Cost of downtime · OEE benchmark · Free OEE tracker · Free OEE software · How to track downtime · How to implement OEE · Light MES vs full MES