Choosing people counting sensors for a hall by the question you need answered

A supplier walks a show director around a hall with a floorplan and a marker, circling doorways. By the end of the walk there are fourteen circles and a quote. Nobody has yet asked what any of the fourteen numbers will be used for.

Choosing people counting sensors is much easier from the other end. Write down the decisions the show office actually takes with a count, then work backwards to the instrument that can answer each one. Most halls turn out to make three separate decisions, and no single instrument answers all three.

Three decisions, and why they need different instruments

The first decision is how many people came. That number leaves the building. It goes to exhibitors, to the board, to the audited attendance figure, and to whoever compares you with the rest of the market. CEIR builds its Index on four components, and attendees is one of them, alongside net square feet, exhibitors and real revenues. CEIR reported in its Q2 2025 Index that attendees sat 3.7 per cent below the equivalent 2019 quarter, net square feet 4.9 per cent below and real revenues 15.6 per cent below, with the total index 8.4 per cent under 2019. Your door number is one of the four inputs to how the market reads a whole sector, which is a reasonable argument for spending real money on it. A door total is a published figure with an audience.

The second decision is where people spent their time. That one stays internal and it drives floorplan, feature placement and next year's pricing conversation. It needs a person to be recognisably the same person in two different places, which is a fundamentally different requirement from a total.

The third decision is whether a space is safe and comfortable right now. Fruin set out the level of service concept for pedestrian spaces in 1971, and the variable he built it on is density, measured in persons per square metre. A headcount cannot express it. A hall containing 6,000 people is fine. Six thousand people in one 400 square metre aisle block is an incident. Grading those density bands is its own subject, and the operational reading of aisle density belongs elsewhere in this cluster.

Three decisions. A total, an identity and a density. Everything about the hardware follows from which of the three you are buying.

What question is each instrument actually able to answer?

Directional counters at a doorway. Overhead sensors that resolve a crossing and its direction, whether they work on stereo vision, time of flight or lidar. They give entries, exits and a running net, anonymously, on every human who passes. They cannot tell you that the person who left at 12:40 is the same person who came back at 13:35, which is why running occupancy from these alone drifts.

Identified badge portals. RFID or barcode reads at zone boundaries. They give you a person, a place and a time, so they support dwell, sequence and repeat visits. Their weakness is read rate. Anybody the portal missed is invisible, and the misses are not random, since badge orientation and body shielding decide them.

Density instruments over an area. Overhead cameras running density estimation, or lidar covering a floor patch. They report persons per square metre across a defined polygon without identifying anybody. They answer the safety question directly, and they are close to useless for the totals question, because a density figure over an area does not integrate into a count of individuals over a day.

Passive radio sensing. Cheap, wide coverage, no cooperation required. It has stopped producing a defensible unique count for reasons outside your control, and the wifi footfall counting limits that follow are severe enough that I would only use it as a relative index.

Precise positioning. Tag-based time of flight, resolving to well under a metre. It answers the identity question at a spatial resolution none of the above can reach, and the anchor and tag economics of UWB positioning for events put it out of reach for a whole audience.

The door decision, and the mistake people make with it

Most organisers already have entrance badge scanning, and most then treat the scan total as the attendance figure. That works only if every human who entered presented a badge to a reader.

They did not. Somebody walked in behind a group. A reader went down for eleven minutes at 09:40. The press door had no scanner. Contractors used the same entrance for twenty minutes before the doors opened. Each of those is a real, ordinary hall event, and none of them appear in the scan file as a gap, because a missing scan looks exactly like a person who did not come.

A directional counter sitting beside the scanner solves this, and it does so in a way that is worth more than either number alone. Suppose the counters at eight public doors record 10,320 entries on day one and the badge system records 9,840 identified arrivals. The ratio, 9,840 over 10,320, is 95.3 per cent, and that is your scan coverage. Report the identified figure as the attendance and the coverage rate beside it, and you have a number with a stated completeness instead of a number with an unstated one.

The counter is doing a different job here from the one the supplier described. It is auditing your primary instrument.

The zone decision needs an identity, so the anonymous options drop out

Dwell time, sequence and repeat visits all require the same person to be recognisable at two moments. Anonymous counters cannot do it at any price.

So zone measurement means portals, and portals mean a decision about coverage. Read in only, and dwell is derived from the next zone entry, which is wrong for anybody who leaves the way they came. Read in and out, and you double the frames per zone and get a real interval.

Take four zones you actually report on. Read in and out and that is eight portal frames, each needing power, mounting and a lane wide enough that people do not walk around it. Plan the read rate honestly at the design stage, because a portal at 96 per cent per crossing sounds excellent and loses a great deal more than 4 per cent of complete journeys once a journey involves several crossings.

Where people get lost between those zones is a related but separate signal, and the search terms and abandoned routes that make up exhibition wayfinding analytics come from the wayfinding tool rather than from any of this hardware.

What does a hall that needs all three end up buying?

Work the shopping list for one hall that genuinely takes all three decisions.

Eight public doors, each needing a directional counter for the total and the coverage audit. Eight units. Four zones reported on for dwell, read in and out, so eight portal frames. One known bottleneck, the pinch where the main aisle narrows past the catering block, needing a density instrument. One unit.

Seventeen devices. Compare that with the supplier's fourteen circles, which were all the same instrument, all at doorways, answering one of the three questions and neither of the others. The seventeen cost more and produce three usable answers instead of one, and every one of the seventeen traces to a named decision.

The arithmetic also tells you what to cut when the budget lands. Drop the density instrument and you keep the total and the dwell and you lose the ability to say anything defensible about crowding at the pinch. Drop the four exit portals and you halve the zone hardware and your dwell figures become estimates with a known lean. Drop the door counters and you keep everything except the ability to state how complete your attendance number is. Those are three different losses, and a show director can choose between them once they are written down that way. Making that choice explicit is most of what good attendee analytics planning consists of.

Where this stops

The decision-first approach has a real weakness, which is that it assumes you know the questions in advance. Some of the most useful things a hall's data tells you are answers to questions nobody asked. A density instrument installed for safety will show you that one aisle empties completely between 15:00 and 15:40, and that observation might change your feature programming, even though nobody specified it.

So there is a genuine argument for a little more coverage than the decision list justifies, and I would take it in one specific form. Instrument one extra zone you have no current reporting need for, chosen because you suspect something about it. Treat it as an experiment with a defined review date and take it out if it teaches you nothing.

The other limit is that none of these instruments is trustworthy on installation. A counter mounted 30 centimetres off its design height, a portal frame with a metal trolley route running past it, or a density camera with a banner hung across its field of view all produce numbers that look completely normal and are completely wrong. Every device on that list of seventeen needs its own check before you report from it, and how to run that check on one door in a morning is a specific exercise in people counter accuracy validation.

The step for this week costs a sheet of paper. List every count your show published or circulated last edition, and beside each one write the decision it changed. Any count with an empty box beside it is hardware you are paying to maintain for no reason, and any decision with no count behind it is the gap worth quoting for.