Fundamentals

Locating staff on board a train

Why station-level location runs out when an officer patrols inside a moving train, and what works there instead: identity rather than coordinate.

Published on 28 August 2026 Metrosafe team

Almost everything written about indoor positioning assumes the space stands still. Platforms, concourses, corridors: fixed geometry, fixed references, a layout that does not change.

On a metro network that describes half the problem. The other half is riding a train.

The stretch nobody talks about

Patrol work on a network does not happen only in stations. A large share of it is on board: the officer boards, walks the consist, gets off two or three stops later, changes train. In terms of hours, it can be most of the shift.

It is also the stretch with the worst operational coverage:

The officer is alone, with no colleagues in sight and no station staff nearby.
On-board incidents — an assault, a disturbance, someone taken ill — are exactly the ones that demand an immediate response.
And it is where the control room has the least information: it knows that officer came on duty, and little else.

The usual workaround is the radio: “where are you?” Which works, except when the person who would have to answer is the one dealing with the problem.

Why station-level ranging does not reach

The base layer of a system like this ranges against the Wi-Fi access points in stations. It works well where those access points are. The moment the officer boards a train, three things change at once and none of them help.

The reference moves

Measuring a distance requires something still to measure against. The station’s access points are fixed; the train is not. Between stations there is simply nothing to range against.

The car is a metal box

A rail vehicle is among the worst possible environments for a radio measurement taken from outside: metal walls that attenuate, parallel surfaces that reflect, and a geometry that changes depending on whether the car is empty or packed. Any measurement that survives that arrives degraded.

And even with a good measurement, you would not know the thing that matters

This is the point that takes longest to see. Suppose the measurement is perfect and says the officer is seven metres from the platform edge. Are they on the platform? Boarding? Already inside the train that is pulling away?

All three give the same coordinate and mean completely different things. The problem stops being one of precision and becomes one of category. No amount of extra metres resolves it.

Changing the question

If the metric position does not answer, ask something else. Not “where are they” but “what are they in”.

The operational answer a control room needs is not a coordinate: it is train 4207, car 3. With that, it already knows where that officer will be in ninety seconds, which platform they will pass through and which door they will appear at.

And that question does have a clean technical answer: one BLE beacon per carriage.

Each car carries a beacon broadcasting its identifier. The officer’s corporate phone detects it and the platform publishes the result, cross-referencing the consist in service to give the train number and the station as well.

Identification, not estimation

The difference between the two layers is not one of accuracy but of kind, and it is worth grasping because it changes how the system gets specified.

| | Station ranging | Car beacon | |---|---|---| | What it returns | A coordinate | An identity | | Has an error margin | Yes, a radius in metres | No | | Can be “almost right” | Yes | No: it either reads the beacon or it does not | | What you specify | Accuracy and latency | Coverage and read rate |

A discrete identification admits no degrees. The phone sees car 3’s beacon or it does not. That removes at a stroke the whole conversation about metres, NLOS, multipath and calibration that dominates metric positioning.

In exchange, other questions appear: how often the beacon transmits, how long the phone takes to confirm a change of car, what happens in the gangway between cars. These are simpler problems with more stable answers.

The obvious objection: wasn’t this supposed to be “no beacons”?

That is the right question, and it deserves a straight answer because it touches the central argument of this kind of solution.

A classic RTLS demands anchors in the infrastructure: cabled, powered devices mounted on platforms and in tunnels, with permits, night-time windows and months of calendar. That is what kills these projects, and that is what is still not needed.

A car beacon is a different animal:

It goes inside the rolling stock, not in the network’s infrastructure.
It is fixed with adhesive: no drilling, no bracket.
It carries its own battery: it connects to nothing on the train, so no electrical type approval is involved.
It is fitted during a depot inspection, not in traffic. No night work on the track.
It is reversible: removing it leaves the car as it was.

And there is an asymmetry that matters more than it looks: people still carry nothing. There are no tags to hand out at the start of a shift, to charge, or to replace when they go missing. That daily process — the one that actually gets abandoned after six months — never appears.

The side effect that often decides the project

There is a consequence that is not obvious and that weighs heavily on the budget in practice.

Mapping an entire network for metric positioning means running radio surveys and calibration in the tunnels too: long stretches, restricted access, scarce work windows. It is the expensive, slow part of the deployment.

With the on-board layer, that work can be avoided. Between stations, car identity replaces the coordinate: you do not need to know which metre of tunnel the train is in, only which train the officer is on and where that service is heading. The radio survey can be limited to whichever stations matter.

Put differently: fitting beacons to carriages can come out cheaper than surveying the tunnels, and it yields more useful information.

When it does not pay off

To put it in the negative too. If patrol work is purely platform and concourse, and nobody patrols on board, this layer is spend without return. The base layer covers the whole case.

It is worth it when at least one of these holds:

A meaningful share of the shift is spent on board.
Coverage per train has to be evidenced for a service contract.
Response to an on-board alert is slow today because nobody knows which train to look in.
You want to avoid the full tunnel radio survey.

And privacy

Locating personnel means processing personal data, here as much as in the base layer, and the same framework applies: purpose bounded to operational safety, prior information to staff and their legal representatives, bounded retention and anonymisation for analytical uses.

It is worth noting that this data is in fact less intrusive than a metric coordinate: what gets published is a car, not a point. That is coarse granularity — enough to coordinate, and far too coarse to watch anyone.

Developed further in what NIS2 requires of a metro operator.