Production counting · Vision on CCTV

Production counting with the cameras you already have

At Grupo Proa, 19 security cameras that were already installed count eggs at six farms with 99 % accuracy measured against manual count, integrated with SAP S/4HANA and in production for more than two years. No special camera or controlled lighting.
Real inference capture: egg trays detected and counted by vision on a security camera
Real inference capture on a security camera

The problem

Why does the production count not match the ERP?

Line10,000Palletizing9,820unrecorded rejectsShipping9,800samplesERP9,600next-day reportFour numbers for the same thing: 400 units unexplained (example)

Counts are done by hand or estimated at close, taken at different points, scrap goes unrecorded and the data reaches the ERP late and keyed in by hand. Every department is right about its own number.

Counting on the line, time-stamped, and sending that number straight to the ERP fixes the first and last causes, and exposes the other two.

The four causes and how to measure them in a week →

How it works

From the security camera to the ERP, in four steps

01Camera02Edge03Count04ERP

Step 1 of 4

The camera already there

An IP security camera that sees the point where product passes. Its video is taken over the network (RTSP or ONVIF), without moving it or changing its recording.

Requirements

What you need to count with your cameras

Almost always, what the plant already has. Anything missing shows up in one site visit.

IP camera with network video

Detail

RTSP or ONVIF. Common security cameras support it

Note

The one already there, if it sees the counting point

A view of the point where product passes

Detail

No permanent obstructions; no special lighting needed

Note

At Proa, the farm camera just as it was

Network between the camera and the edge

Detail

Cable or a stable link inside the plant

Note

No internet needed to count

A place for the industrial computer

Detail

Rack, panel or electrical room with power

Note

One can serve several cameras

A reference to validate against

Detail

Manual count, shipping note or machine counter, for one or two weeks

Note

That is how Proa's 99 % was measured

What it does

  • Counts units per line, hour and shift
  • Detects when the line stopped because product stopped passing
  • Delivers the number to the ERP, the app and an API
  • Keeps counting with no internet

What it does not promise

  • Detecting defects without a labeled image set of your product
  • Counting what the camera cannot see
  • Deciding for you: it delivers the number with its source

Proof

Grupo Proa: the data that arrived a week late

Poultry farm with security cameras above the egg collection lines

99 %

measured against manual count

19

cameras already in place

6

farms

2+ years

in production

They counted by hand at 17 points, with 5 % to 15 % deviation, and matching the count to the batch took a week. Today the count reaches SAP in real time.

Read the full case →

In the app

What the shift sees when the camera counts

The app is free during early access. Camera counting requires the edge in the plant.

iGromi app: alert for packaging line 2 stopped, detected by the camera that counts the jars

Your data

Video never leaves the plant

Counting runs on the edge, inside the plant. What goes up to the cloud depends on the plan you choose.

Free

iGromi cloud, separated per company

The shift app: chat, photo and voice records, and AI that answers with its source.

Premium

A private, isolated cloud for your company

Adds SAP, ERP and spreadsheets. The AI model runs inside that cloud.

Enterprise

Inside your plant, on Edge or on-premise

Adds PLCs, sensors and cameras. The Edge reads the PLC read-only and keeps measuring offline.

FAQ

Camera counting FAQ

Does it work with any security camera?
With most IP cameras, because they stream video over the network via RTSP or ONVIF. What matters is not the brand but the view: the camera has to see the point where product passes. If it does not, it is repositioned or a camera is added at that point.
How accurate is counting with existing cameras?
At Grupo Proa, 99 % measured against manual count and validated with the customer, on 19 security cameras already installed across six farms. In a new plant accuracy is measured the same way, against the reference the operation already uses, before the system is signed off.
Why does the production count not match the ERP?
For four compounding reasons: counting is done by hand or estimated at close; it is counted at different points (machine, palletizing, shipping); scrap and rework are not recorded; and the data reaches the ERP late, aggregated and keyed in by someone else. Counting on the line with a time stamp and sending that number straight to the ERP fixes the first and last, and exposes the other two.
How does it avoid counting the same unit twice?
Each unit is tracked while it is in frame and counted only once, when it crosses a virtual line defined in the image. If it stops, moves back or is briefly hidden, it does not add again.
Can it count by product and by shift?
By line, hour and shift, always. By product, when products can be told apart visually or when the count is matched with the active production order.
Does the video leave the plant?
No. Video is processed on an industrial computer inside the plant and never leaves it; only the counts travel. If the internet drops, it keeps counting and sends what is pending when the network is back.
Does it integrate with SAP or another ERP?
Yes. At Grupo Proa the counts reach SAP S/4HANA in real time. For other systems they are delivered via API or whatever integration the ERP supports.
Does it detect quality defects?
This page is about counting. Detecting defects requires a labeled image set of your product, and we do not promise it without one. For inspection, see the industrial computer vision page.

Every shift without a record is data you lose. Start free.

  1. 1Request early access, free
  2. 2Pick your trade: maintenance, production or safety
  3. 3Your team starts with the chat and its trade helper

No credit card. Your data stays yours.

PLANT · SHIFT B
Tasks
Luis · maintenance
OVERDUE
Sugar silo 1
No signal since 05:40
Sensor · supply
TODAY
M-204 vibration
4.1 mm/s, above normal
Detected by the sensor
PROPOSED
Check the capper
14 badly seated lids since 08:30
AcceptDismiss
HomeTasksChatMore