Skip to main content

RUBICON

Capability Reference

What A-Eye can verify, and what it cannot

A working reference for engineers and operations teams evaluating whether camera based verification can answer their question.

Why this page exists

Most vision projects fail because the question asked was never answerable from the camera that was available. It is cheaper to establish that before a deployment than after one, so the limits are published alongside the capabilities.

Capability matrix

Status of each verification type.

Person presence
Detects that a person is present in a zone, and for how long. Does not identify who.
Available
Production counting
Counts discrete items crossing a defined line or zone, by class where classes are visually distinct.
Available
Pallet and carton counting
Counts palletised and boxed goods in racking, staging areas and dock bays.
Available
Vehicle and forklift detection
Detects vehicles and forklifts, their movement, and dwell time in a zone.
Available
PPE compliance
Checks for helmet and high visibility vest on detected people. Smaller items such as gloves and eye protection are far less reliable at typical camera distances.
Conditional
Zone and exclusion breach
Flags entry into a defined area, including person and vehicle in the same zone.
Available
Stock movement
Records goods moving in, out and between zones, reconciled against expected movements.
Available
Number plate reading
Readable where the camera is positioned for plates specifically, at a suitable angle and distance. A general area camera is usually not sufficient.
Conditional
Face recognition
Technically evaluated and working, but requires enrolment, carries privacy obligations, and the models assessed have licensing constraints for commercial use. Treated as a separate decision, not a default.
Constrained
Workforce presence intelligence
Continuous presence measured against expectation rather than door events.
In development
Production analytics (OEE, downtime)
Analytics layer built on verified counts. Counting itself is available today.
In development
Digital twin
A live structured model of the site beyond the current site plan.
Roadmap
High speed defect inspection
Sub millimetre inspection on a fast moving line is a different discipline requiring dedicated optics and lighting. Specialist inspection vendors are the right fit.
Not supported
Reading small printed text
Batch codes and small labels are generally unreadable from ceiling mounted general area cameras.
Not supported
Intent or emotion inference
We do not infer what a person intended or how they felt. It is not reliable and we will not present it as evidence.
Not supported

Conditional means it depends on camera placement, resolution and scene conditions. We establish which during assessment using your own footage.

What changes the answer

Four things decide whether a question is answerable.

1. Pixels on target

The single biggest factor. What matters is not camera megapixels but how many pixels land on the object you care about. A 4K camera covering a whole warehouse may give a smaller target than a 1080p camera covering one aisle.

2. Angle

A camera looking along a line counts poorly because objects occlude each other. The same camera moved to look across the line counts well. Placement is usually a bigger lever than hardware.

3. Occlusion

If people, racking or vehicles regularly block the view of the thing being counted, results degrade. Persistent occlusion is a reason to move a camera rather than accept poor data.

4. Lighting and consistency

Verification tolerates ordinary industrial lighting. It does not tolerate a scene that swings between direct sun and deep shadow across a shift, or areas that go effectively dark without infrared.

Self assessment

Is your question likely answerable?

Usually yes

  • The thing you want counted is visible to a human watching that camera
  • It is bigger than roughly a shoebox in the frame
  • It crosses a predictable point rather than appearing anywhere
  • The area is lit consistently through the shift
  • You want a count, a duration, a presence or a movement

Usually no, or needs a different camera

  • A person could not reliably tell from that footage either
  • It depends on reading small text or fine surface detail
  • The object is usually hidden behind something else
  • It requires identifying a specific individual by face
  • It requires judging why someone did something

A useful test: if you could not answer the question yourself by watching that camera feed, the software will not either. The value is that it watches every camera continuously and keeps the evidence, not that it sees things people cannot.

Common questions

Why publish what the system cannot do?

Because the alternative wastes everyone time. Most failed vision projects fail because the question asked was never answerable from the camera that was available. Knowing that at the assessment stage is cheaper than finding out after deployment.

Can A-Eye identify specific individuals?

Person detection and presence are standard and do not identify anyone. Identifying named individuals requires face recognition, which we treat separately: it needs explicit enrolment, raises privacy obligations, and the models we have evaluated carry licensing constraints for commercial use. We will not deploy it casually and we will tell you if it is the wrong tool.

How accurate is detection?

Accuracy is a property of a specific camera watching a specific scene, not of the software. A pallet at eight metres in good light and a pallet half occluded behind a forklift are different problems. We measure accuracy on your footage during assessment rather than quoting a headline number that would not survive your site.

Can it work with our older analogue cameras?

If they are connected to a DVR or encoder that exposes an RTSP stream, usually yes. Image quality sets the ceiling on what can be verified, so older or lower resolution cameras narrow the list of answerable questions rather than ruling everything out.

What happens when the system is unsure?

It is designed to record uncertainty rather than guess. An event that cannot be verified confidently is surfaced as unverified rather than being counted as a fact. Numbers that are presented as verified are the ones backed by evidence.

Find out what your cameras can actually verify

An operational assessment looks at your cameras, your zones and your questions, then tells you which are answerable today and which are not.