Can Computer Vision Count Objects on a Moving Conveyor?

Yes, computer vision can count objects on a moving conveyor. The real question is whether the setup can count them reliably at your line speed.

Counting sounds simple until production starts moving. Products touch each other. Packaging reflects light. Operators change spacing. Conveyors vibrate. Some items tilt. Labels face different directions. Dust settles on the lens. A batch change introduces a slightly different shape.

A human standing near the conveyor may adjust naturally. A computer vision system needs the right camera view, lighting, trigger logic, software model, and validation rules.

When designed properly, computer vision can count objects on moving conveyors for production tracking, packing verification, rejection control, dispatch accuracy, and loss reduction. But it must be designed around the actual movement of the line.

How conveyor counting works

A camera is mounted above or beside the conveyor. The system captures images or video frames as items pass through a defined inspection zone. The software identifies each object, tracks it across frames, and increments the count only when the object crosses a virtual line or meets a counting condition.

In some setups, the camera works with sensors or PLC signals. For example, a photoelectric sensor may trigger image capture when an item enters the zone. In other setups, the vision software continuously watches the conveyor and tracks objects directly.

The best approach depends on speed, spacing, object size, lighting, and the accuracy required.

What makes counting easier

Counting is easier when objects are clearly separated, visually consistent, and moving in a predictable path. A row of cartons with regular spacing is much simpler than loose parts touching each other randomly.

Accuracy improves when:

• Products have visible boundaries
• Lighting is stable
• Conveyor speed is consistent
• Objects do not overlap heavily
• Camera angle is fixed and protected
• The background contrasts with the object
• Product variants are known in advance
• The counting zone is physically controlled

This is why mechanical discipline still matters. Computer vision is powerful, but a chaotic conveyor creates avoidable errors.

What makes counting difficult

The most common problems are overlap, reflection, high speed, inconsistent placement, and object similarity.

For example, transparent bottles can be difficult because edges are not always visible. Shiny pouches may reflect overhead lights. Small parts on a dark conveyor may blend into the background. Products touching each other may look like one object. Items moving too fast may blur without proper shutter and lighting.

A good implementation does not ignore these issues. It solves them through camera selection, lens choice, lighting design, mounting stability, conveyor guides, and software logic.

Counting is not the same as inspection

A system may be able to count objects without inspecting their quality. Another system may count and inspect at the same time.

For example, on a packing line, computer vision can count the number of sachets entering a carton. On another line, it can count bottles and also check cap presence. On a third line, it can count components and verify orientation.

The more tasks you add, the more carefully the system must be designed. Counting alone may need one camera. Counting plus label reading, defect detection, and orientation checking may need multiple views, lighting zones, and integrated decision logic.

Where manufacturers use conveyor counting

Computer vision counting is useful in many production situations:

• Counting finished goods before packing
• Verifying pieces in a carton, tray, or kit
• Tracking output per shift or machine
• Detecting missing items on a line
• Counting rejected parts separately from accepted parts
• Matching production count with ERP or dispatch records
• Monitoring line speed and stoppage patterns
• Preventing underfilled or overfilled packs

The biggest benefit is not just count accuracy. It is the visibility created when counts are connected to production records.

A standalone counter may tell you 4,820 items passed. A connected system can tell you which SKU, batch, machine, shift, order, rejection reason, and dispatch plan those items belong to. That is where AICAN Optiwise can help manufacturing teams convert vision counts into operational decisions.

How accurate can vision counting be?

Accuracy depends on the use case. In controlled setups, vision counting can be highly reliable. But no responsible partner should promise a universal accuracy number without seeing the conveyor, product, speed, spacing, lighting, and error tolerance.

Instead of asking, "What accuracy do you guarantee?" ask:

• What are the likely counting error scenarios?
• How will the system handle overlapping objects?
• How will we validate counts during the first month?
• Can the system flag uncertain counts?
• Can manual correction be recorded with reason codes?
• How are counts reconciled with production or dispatch data?

This leads to a more useful conversation.

Conveyor speed changes the design

A slow conveyor may work with standard cameras and simple image processing. A high-speed line may require industrial cameras, controlled lighting, short exposure, hardware triggers, and edge processing close to the machine.

If the line speed varies, the system must account for it. Otherwise the same product may appear across multiple frames in a way that causes double counting or missed counting.

This is why integration with conveyor controls or PLC signals can be valuable. The vision system should understand when the line starts, stops, slows, or reverses.

What about mixed products?

Computer vision can count mixed products if the system is trained or configured to distinguish them. For example, it may count red and blue packs separately, identify different bottle sizes, or classify components by shape.

However, mixed-product counting needs better data discipline. The system should know which SKU or batch is running, and operators should have a clear method for product changeover. If the wrong recipe is active, counts may still happen, but classification may be wrong.

Connected production systems reduce that risk by linking the running order, item master, and inspection recipe.

What should happen after counting?

Counting is only useful if the result triggers the right action.

A conveyor counting system may:

• Update production quantity automatically
• Stop the line when count mismatch occurs
• Trigger a rejection mechanism
• Alert the operator when a pack is incomplete
• Send accepted and rejected counts to a dashboard
• Support dispatch verification
• Create batch-wise traceability

For many factories, the real ROI comes when counting reduces manual entry and prevents mismatch between physical production and system records.

AICAN builds around this connected view of manufacturing: production should not live in one place, quality in another, inventory in another, and dispatch in another. They should speak to each other.

Where AICAN Optiwise fits

With AICAN Optiwise, conveyor counts can become part of the wider manufacturing workflow. Accepted count, rejected count, batch quantity, shift output, stock movement, and dispatch readiness can be aligned instead of manually reconciled later.

That helps plant teams answer practical questions faster: Did we produce what we planned? Did rejects increase on one line? Is packing short? Is dispatch relying on correct quantity? Which shift needs attention?

You can read more about the company at About AICAN.

Founder's Note

Counting is one of those factory problems that looks small from outside and becomes painful inside operations. A few missed pieces can create customer complaints, dispatch mismatch, rework, and arguments between production and inventory.

Computer vision helps when it is treated as part of the process, not just a camera above a conveyor. The count must connect to action.

FAQs

1. Can computer vision count very small parts?

Yes, but small parts need careful camera resolution, lens choice, lighting, and separation. If parts overlap heavily, mechanical separation may be required.

2. Can it count products at high speed?

Yes, if the system uses suitable cameras, lighting, triggering, and processing. High speed should be tested with real conveyor conditions before rollout.

3. Can one camera count multiple lanes?

Sometimes. It depends on lane width, product size, camera resolution, and whether products overlap visually. Multiple cameras may be more reliable for wider or complex lines.

4. Can counting data go into ERP or dashboards?

Yes. Vision counts can be integrated with production, inventory, quality, and dispatch systems. That is where the business value becomes stronger.

5. What is the biggest reason conveyor counting fails?

Poor control of the physical setup: unstable lighting, moving cameras, overlapping products, inconsistent spacing, and unclear action rules after a mismatch.