Updated July 2026 · 6 min read · Adente Vision Engineering Team
What does the Counting inspection mode do?
The Counting mode answers a plain question: is the right number of things present. It runs on the same unit as the other checks and turns each captured frame into a verified quantity, parts crossing a belt, cavities filled in a mould, items in a blister or a carton, connectors on a harness. Rather than an operator watching and tallying, the unit counts every part and hands the number to the line.
Counting is one of four inspection modes on the unit: Anomaly, Defect, Counting and Quality. Adente Vision is an edge-AI visual inspection unit built by ADENTE Advanced Engineering Technologies, part of the Aden Group, sold through automation system integrators, and the Counting mode is the one you reach for when the defect is a wrong quantity rather than a wrong appearance. It uses the deterministic side of the hybrid approach, classical computer vision locating and counting discrete items, so the number is repeatable rather than a probability.
Because it is a mode on a unit already inspecting the part, counting does not need a second camera or a separate sensor bank. The same capture that judges quality can also return the count.
Why count on the line instead of a manual tally?
A manual tally is a sample, not a guarantee. An operator checks some packs, some cartons, some cycles, and infers the rest, so a short-count that lands between checks reaches the customer. On-line counting inverts that: every part is counted, so a wrong quantity is caught as it happens rather than discovered in a complaint or an audit weeks later. This is the difference between acceptance sampling, formalised in standards like ISO 2859-1, and 100% verification at line speed.
There is a cost difference too. A short-shipped carton or a missing component in an assembly is expensive to correct downstream and worse to ship, while an extra part is waste. Counting every part on the line makes both visible at the point they occur, where the fix is cheap. The operator moves from tallying to handling the exceptions the unit flags, which is a better use of the person than counting.
How does counting keep up at line speed?
Counting has to be per-part and immediate, or it is just sampling with extra steps. The unit decides on the edge: the measured field result is about 30 milliseconds per part, and the conservative catalog bound is 0.5 seconds per part at a throughput of 100+ parts per minute. Treat these as the envelope; the count-rate you can commit to for your own parts, spacing and lighting needs an application-specific measurement.
The table below sets on-line counting against the manual tally it replaces.
| Aspect | Manual tally | Counting inspection mode |
|---|---|---|
| Coverage | A sample or spot check | Every part, 100% at line speed |
| Speed | Limited by the operator | 100+ parts per minute, about 30 ms per part |
| Record | Handwritten, easy to lose or mis-enter | Logged to the web dashboard, sent to the PLC |
| Combined check | Count only | Count plus pass/fail in a single pass |
| Reaction | After the fact | Real-time signal to the PLC to stop or divert |
Capture locks to the part, not to a clock: the unit triggers off an encoder pulse, a photoelectric sensor or a fixed interval, so the count stays aligned to the flow even when line speed varies.
How do you combine a count with a pass/fail check?
The strongest use of counting is not counting alone; it is the right number and all of them good, in one pass. Because the same capture can run more than one check, the unit can confirm the quantity and judge the parts at once: a carton with the correct count but one cracked item still fails, and a blister with all-good items but one empty pocket still fails. The verdict the PLC receives is a combined answer, not two separate readings taken at two stations.
That pairing is what makes counting useful for packaging and assembly, where completeness and condition are both requirements. The unit returns pass/fail and a reject reason, so an operator reading a reject sees whether it was a short count, an over-count or a bad part, and the reject reason travels on the fieldbus word alongside the discrete pass/fail bit.
Where does counting overlap presence and absence?
Counting and presence/absence are neighbours, and it helps to know which to reach for. Presence/absence asks a yes/no per location, is the clip, screw, label or cap there, and is learned from good assemblies. Counting asks how many across the frame, and reports a quantity. On a fixed assembly with known positions, presence/absence is the natural check; on a flow of loose items or a variable count, counting is. Many lines use both: presence/absence confirms each mandated component is in its place, and counting confirms the total.
For the presence and assembly-verification side learned from about 20 good images, see the sibling post on assembly verification; for how the mode fits the wider set of checks, see the pillar guide on AI visual inspection, and browse the real applications to see where counting runs in production.