AI production counting with cameras: cartons, pieces and QC counts
By Graeon.ai · Updated
AI production counting uses a camera above a packing line, loading point or QC station to detect each carton or piece as it passes and keep a running total. The count is written into the ERP against the order, batch or production entry, so dispatch and stock figures come from what the camera saw instead of a clipboard. It works well where items move one at a time through a clear, well-lit view.
Why plants count with cameras
Most plants still count by hand at the end of the line. A packer ticks a sheet, a supervisor totals it at shift end, and someone types it into the ERP the next morning. Each step can slip. Short packing shows up as a customer complaint, and excess packing shows up as a stock gap at month end.
A camera that counts at the point of work removes the tally sheet and the re-typing. The count exists the moment the carton is placed, and it can be checked against what the order says should be there.
Where camera counting is used
- Packing: cartons counted as they are placed on a pallet or pass a packing station
- Loading: cartons or bags counted as they go onto a truck, checked against the dispatch quantity
- QC stations: pieces approved and rejected at an inspection point counted separately
- Inspection lines: items inspected counted alongside whether the operator was present
- Material movement: pieces moved between zones, so work-in-progress figures match the floor
How the count reaches the ERP
The camera only sees objects. The ERP decides what they mean. So the setup has to tell the system which order, batch or shift a count belongs to. The common ways are a supervisor selecting the active order on a tablet at the station, a production plan that fixes which batch runs on which line and when, or a dispatch document opened before loading starts.
Once the count is linked, it can update the production entry, the finished-goods stock or the dispatch register. A mismatch between the counted and the planned quantity becomes an exception for someone to check, not a number found weeks later.
Where camera counting struggles
- Occlusion: a packer's arm or body hides a carton as it is placed, or one item passes behind another
- Stacking: cartons stacked several high, or loose pieces piled in a bin, cannot all be seen from one angle
- Lighting: glare from a shed roof, low light on a night shift or a truck bay facing the sun changes how items look
- Look-alike items: different products in identical cartons need a label, a separate line or an order link to tell apart
- Speed and spacing: items that touch or overlap on a fast conveyor are harder to separate than items with a gap
- Camera movement: a camera knocked out of position by a forklift or vibration stops counting the right area
How to set it up so the numbers hold
- Mount the camera above the counting point, looking down, so items do not hide each other
- Define a counting line or zone that every item must cross once
- Light the station evenly and keep it the same across shifts
- Train the model on your own cartons and pieces, in your real packing pattern
- Run camera counts next to manual counts for a few weeks and investigate every gap
- Agree who reviews mismatches and how fast, before the counts go into the ERP
How Graeon Netra does it
Graeon Netra runs on the plant's existing CCTV. In Netra's footage from a warehouse packing station, each box is detected as it is placed on the pallet and a running total is shown on the feed. At a QC station, Netra records whether the inspector is present and working across the shift, and at a bright bar inspection station Netra counts bars inspected while also flagging when the station is left unmanned.
Models are trained on the site's own cartons, and counts are recorded against the order. Because Graeon builds both Netra and Graeon ERP, a count at packing becomes a production or dispatch entry without re-typing. Graeon ERP has ready connectors for Tally and Oracle; other systems are connected through the API as a project.