Who this is for
Plant owners, quality managers, and engineering leads on food, beverage, plastics, packaging, pharmaceutical-adjacent, and metal-fabrication lines who are losing money or customers to defects that manual inspection misses — and want to understand what automated vision inspection costs and where it actually pays.
What machine vision inspection is
A machine vision inspection system is an industrial camera (or several), purpose-chosen lighting, and image-analysis software that checks every product as it passes on the line. It is often called automated optical inspection (AOI). Where a human inspector samples a fraction of output, gets tired, and disagrees with the next inspector, a vision system checks 100% of production against identical criteria in milliseconds, logs every result, and triggers a reject when something is wrong.
It is not a camera with a recording. It is a decision-making system: good part or bad part, pass or fail, on every unit, every shift.
Decision rule: if a visible defect is reaching your customers, or manual inspection is expensive, inconsistent, or a bottleneck, vision inspection is worth costing. If the defect isn't visible, or you already catch it reliably and cheaply, it may not be.
AOI vs manual inspection
| Dimension | Manual inspection | Machine vision (AOI) |
|---|---|---|
| Coverage | Sample (a fraction of output) | 100% of production |
| Consistency | Varies by person, shift, fatigue | Identical criteria every time |
| Speed | Limited; can bottleneck the line | Line speed, milliseconds per item |
| Record / traceability | Rare, manual | Every result logged automatically |
| Fine / fast defects | Easily missed | Caught reliably if visible |
| Running cost | Ongoing labour, every shift | Capex up front, low running cost |
| Best at | Judgement, rare/varied edge cases | Repetitive, defined, high-volume checks |
The two are not mutually exclusive. The best setups use vision for the repetitive, high-volume, well-defined checks and keep human judgement for the rare and ambiguous. Vision removes the dull 95% so people can focus on the 5% that needs a brain.
What it costs in South Africa, 2026
Cost is driven by camera count, line speed, lighting complexity, and algorithm difficulty — not by a price list. Indicative bands:
| System | Indicative cost (ZAR) | Typical use |
|---|---|---|
| Single-camera station | 120 000–300 000 | One check — code verification, label presence, a single defect type |
| Multi-camera inline station | 350 000–900 000 | Several checks at once, custom algorithms, full line integration and reject |
| High-speed / multi-station system | 1 million+ | High line speeds, 360° inspection, multiple inspection points, complex defects |
For context: a single label or code defect that triggers a retailer penalty, a recall, or a rejected export consignment can cost more than a whole inspection station. That is why the payback question matters more than the sticker price — see total cost of ownership for the same logic applied to a whole line.
Where vision inspection pays fastest
Not every line needs it. It pays fastest in these situations:
- A defect is reaching customers. Complaints, returns, or rejected consignments mean the cost is already being paid — in money and reputation. Vision moves the catch point back onto your floor.
- Manual inspection is a bottleneck or expensive. If inspectors slow the line or you run several per shift, the labour cost and throughput drag often justify automation quickly.
- A missed code or label triggers a penalty. Wrong/missing date codes, lot numbers, or labels can mean recalls, retailer fines, or compliance failures. Code verification (OCV) is one of the highest-payback checks there is.
- The defect is fine, fast, or subtle. Hairline cracks, small contamination, fill-level deviation — things humans miss at line speed but a camera catches every time.
- You need traceability. Regulated or export markets increasingly require proof that product was inspected. Vision logs every result automatically.
The checks with the best payback, by industry
- Food & beverage: fill-level, cap presence/seating, label position and SKU, date/lot code verification, seal integrity. See food & beverage lines.
- Plastics & packaging: moulded-part defects (short-shot, flash, voids, colour), print and label inspection, dimensional checks. See plastics & packaging lines.
- Metal fabrication: surface and weld defects, dimensional measurement, presence and assembly verification. See metal fabrication lines.
- Agro-processing & general: foreign-material and contamination detection, pack count and completeness, code verification before dispatch.
Failure mode: buying a cheap "smart camera" appliance off a spec sheet, pointing it at the line, and expecting it to work. Reliable inspection comes from lighting design, the right algorithm, and tuning on real product — not from the camera's megapixels.
Why lighting and algorithm — not the camera — decide success
The single biggest reason vision projects fail is poor imaging. Get the lighting wrong and the best camera and algorithm in the world cannot reliably tell a defect from a shadow. Good machine vision engineering spends most of its effort on:
- Lighting — the geometry, colour, and type (backlight, dome, dark-field, coaxial) that makes the defect obvious and the background irrelevant. This is 80% of reliability.
- The algorithm — logic tuned to your product's natural variation, so it catches real defects without rejecting good product. A check that flags 5% of good product as bad will be switched off by the operators within a week.
- Tuning on real production — validating against a representative set of real good and bad samples, then driving false-reject and missed-defect rates to agreed targets before sign-off.
This is exactly why CISH designs the inspection algorithm rather than reselling a fixed appliance — we build the check on the Hikvision VisionMaster platform and tune it on your product.
How to scope a system that works
- Define the defect from real good and bad samples, and the acceptable false-reject rate.
- Confirm the line speed the inspection must run at.
- Design the imaging and lighting for your product and defect.
- Build and validate the algorithm against representative samples.
- Integrate and reject — interface to the line PLC, add a reject mechanism, log results.
- Tune on real production until false-reject and missed-defect rates hit target.
Vision inspection and your OEE data
Inspection results are quality data — and quality is one of the three pillars of OEE (availability × performance × quality). A vision station that logs every pass/fail feeds directly into your OEE picture and surfaces where and when defects spike. This is why vision inspection sits naturally inside a wider line upgrade and digitalisation programme rather than as an island. See measuring OEE on an old line for the foundation it plugs into.
What CISH does
We design the inspection algorithm for your product and defect, choose the camera and lighting, build the station on Hikvision VisionMaster, integrate it with your line PLC and reject mechanism, and tune it on your real production. See Machine Vision Inspection for how we engage, and our technology partners for the platform stack.
Frequently asked questions
What is the difference between machine vision and AOI?
They're effectively the same thing in this context. "Machine vision" is the broad technology; "automated optical inspection (AOI)" is the application of it to inspecting products on a line. Both mean cameras plus lighting plus software making pass/fail decisions automatically.
How much does a machine vision inspection system cost?
In South Africa in 2026, a single-camera station typically starts around ZAR 120 000–300 000, a multi-camera inline station ZAR 350 000–900 000, and high-speed or multi-station systems exceed ZAR 1 million. Cost depends on camera count, line speed, lighting, and algorithm complexity.
Can vision inspection be added to an old line?
Yes. A station is usually a self-contained module — cameras, lighting, reject — interfaced to the line PLC. It does not require replacing your line controls and is a natural part of a line-upgrade project.
Will it slow my line down with false rejects?
Only if it's badly built. Reliability comes from lighting design, the algorithm, and tuning on real product. A well-engineered station is commissioned against agreed false-reject and missed-defect targets, so it improves quality without throttling throughput.
What can it actually inspect?
Defects and surface flaws, dimensions, printed codes and dates (OCR/OCV), labels and print quality, fill level, cap and seal integrity, presence/absence, and count. Because the algorithms are custom-designed, almost any visible product characteristic can be checked.