Velosyti

Vision AI for counting and packing

Counting pieces by hand at the end of a packing line does not scale, and the errors are expensive. A camera can do it, if it is set up properly.

· 2 min read · For factory and warehouse managers

Somewhere near the end of most packing lines, a person is still counting pieces by eye before a carton is sealed. It works, until volumes rise or shifts get tired, and then miscounts start costing you — short shipments that bring complaints, or overpacking that quietly eats margin. A camera that counts consistently, every time, removes this specific failure point.

What vision-based counting actually does

  • Watches items pass on a conveyor or under a fixed camera and counts them as they move.
  • Flags a carton that does not match its expected count before it is sealed, not after a customer complains.
  • Distinguishes between similar-looking items — sizes, colours, variants — that a rushed manual count might mix up.
  • Keeps a record of counts per batch, useful when a dispute over a shipment comes up later.

What it needs to actually work

Lighting and camera placement matter more than the AI model itself — a camera that cannot see items clearly will miscount no matter how good the software is. Items that overlap or move too fast for the camera's frame rate need either a slower line at that point or a better camera, not a smarter algorithm alone. This is worth getting a vendor to assess on your actual line, not just quote from a spec sheet.

What it costs you

Camera and mounting hardware is a one-time cost per counting point; the software is ongoing but usually small compared to the labour and error cost it replaces. It pays off fastest on high-volume, repetitive counting points — not on low-volume, highly varied packing that a person can judge better than a camera anyway.

Questions to ask any vendor

  • Have they tested this on your actual product, under your actual lighting, or only on a demo sample?
  • What is the accuracy at your real line speed, not a slowed-down test?
  • Does it flag a mismatch before the carton is sealed, or only report errors afterward?
  • Can it tell apart your specific product variants, or just count generic objects?

How we can help

We build vision-based counting and inspection using Vizhi, our own vision model, tuned to your specific product and camera setup rather than a generic off-the-shelf model. See our factory work, or talk to us to get your line assessed.

References

  1. Computer vision — overview
  2. GS1 — global standards organisation, overview