People counting technologies compared: beam, thermal, 3D stereo, time-of-flight, AI video and Wi-Fi (2026)

Accuracy figures are the ranges vendors publish for their own products under good conditions; the two sensors we sell are marked. Mounting heights are typical published ranges. “Privacy” describes what the sensor can capture by design, which is what matters legally, not what a particular installation is configured to keep.

An accuracy rating is measured under particular conditions. What matters is whether it holds on every other kind of day, because the point of counting is comparison: this Saturday against last, this year against last. A sensor rated at 98% that loses count whenever the entrance is sunny is 98% most of the time and far less for those hours, and once one period is measured differently from another, the comparison is gone. A counter that misses the same small share of people every day, by contrast, still tells you the truth about your trends.

So the useful question is not which technology has the shortest list of weaknesses, but which of them your door will trigger. Beams are thrown by a door that swings through the beam or a handle that sticks into it. Thermal can be thrown by heat sources and by people who stand still. Stereo needs a little light. Some time-of-flight sensors are thrown by direct sunlight on the entrance floor. Phone-signal sensing counts devices, not people, however well it works. Match the weaknesses to the entrance and most accuracy problems never happen.

The other half of consistency is not missing time. The more devices that sit between the door and the report — a camera and its recorder, a sensor and its gateway or cellular box — the more things can stop the counting, and the same goes for a network outage on a sensor that does not store its counts, or batteries nobody replaced. A counter that records nothing for an afternoon has not made a small error; it has made a hole. The more that can go wrong, the less you can trust the number. PEARL and the 3D Scope II LC are both designed around these: they store counts through an outage and connect straight to your network; PEARL warns you before its batteries run low and keeps counting in direct sun on most single and double doors; and the 3D Scope II LC’s HDR sensor handles bright, backlit entrances. Our accuracy guide sets out each one.

The original people counter, and still in wide use for plain reasons. An infrared transmitter on one side of the doorway shines at a receiver on the other; when a body interrupts the beam, the receiver registers a count. Beams used to be wired, and running a cable to each side of a doorway often cost more than the counter itself; wired beams make little sense today. Wireless units removed that cost, which is why beams are still the default for a standard door.

Variants

Reflector beams use one active unit and a mirror; cheaper, but the reflector is bulky, easily knocked and prone to sunlight errors. Transmitter/receiver (Rx/Tx) beams are smaller and more reliable, and are what PEARL uses. Directional beams put two beams a few centimeters apart on each side: A then B is an entry, B then A an exit. Wireless beams pulse the beam to save power and run on batteries for most of a year; they are the entry point for most first-time buyers, at the price of a battery routine — and because PEARL stores counts through a network outage and emails you before its batteries run low, a dropped connection or a late battery change costs you little. For a chain of a hundred locations, that maintenance load is the real argument for spending more.

Verdict: for single and double doors that open outward, a beam is the simplest and cheapest counter to run and to check. Other technologies are better for wide entrances, but a beam’s range and low cost still make it worth considering there.

PEARL: an Rx/Tx pair on AA batteries over 2.4 GHz Wi-Fi, peel-and-stick, in/out, 15 ft range (7 ft in bright sun, which still covers most single and double doors), upwards of 98% on adults walking through normally, one at a time, $499.95 with software from $0, typically 6–9 months on a set of AA batteries. It is the right sensor for a single or double door, and the wrong one for a wide and very busy entrance — which is why we also offer a 3D stereo sensor.

A low-resolution thermal camera looks straight down and sees people as warm blobs moving across a cooler floor. Firmware tracks each blob and counts it when it crosses the line. Thermal was pioneered in the UK in the 2000s by Irisys, which stopped selling hardware in 2023 and has since closed.

Variants

Thermal still comes in wired form, and is also the technology behind today’s battery overhead counters, which pair the thermal array with machine learning and connect over Wi-Fi, with cellular as an option.

Verdict: for a dark indoor entrance thermal is a good choice, and battery models can be fully wireless. Look closely at the cost: some are sold only with a paid software plan, which can make five years of ownership cost more than many 3D counters — about two and a half times the 3D Scope II LC on its full analytics plan. Anywhere with light, 3D stereo remains the technology to beat.

A single camera looks down and software decides what crossed the line. How well it does that depends almost entirely on the software, and 2D has changed more in the last few years than any other technology here.

Variants

Background-subtraction units, the older kind, separate moving foreground from a learned background, which is where 2D’s reputation for shadows and shopping carts comes from: a shadow moving across the floor, a cart, a child and an adult were all just moving pixels. On-device AI units recognize a person by appearance, and the difference is large; what “AI” actually means on a 2026 datasheet is below. Re-used security cameras are a third option, and a different proposition:

A security camera is mounted at an angle to see faces, which is the wrong geometry for separating people who overlap. It is also doing another job: the counting depends on a device chosen for surveillance, a separate NVR or DVR doing the processing, and a data path from camera to recorder to platform — every stage another thing that can fail or fall behind, where a dedicated counter does the work where it stands.

The reasonable answer to that is “so add a camera, pointed straight down”. It is a fair point, and it is also the moment the comparison changes. You are now buying hardware, choosing a model suited to that entrance’s light and ceiling height, and paying someone to mount and aim it — the same work a purpose-built counter needs — while still keeping the recorder, the integration and every link in that chain. Once you are installing something either way, a dedicated counter is competing on equal terms, and the analytics license for the camera route usually costs more than the sensor would have.

Verdict: a modern 2D counter with on-device AI, mounted straight down, is a serious option and closing the gap on depth sensors quickly; depth still has the edge on close crowds and hard shadows, because it measures rather than infers. A re-used security camera is a different purchase, to be judged on the whole chain rather than the camera.

A stereo counter sees depth the way your eyes do. Each eye sees the scene from a slightly different angle, and your brain turns the difference into distance; the sensor does the same with two lenses a few centimeters apart, and works out how tall everything below it is. That is what a single camera lacks. An adult is a head at adult height, a child is shorter, a cart is low and flat — and on-device AI adds shape recognition on top of the measurement, so a cart or a wheelchair is told apart from a person. Each person is tracked across the view and counted once, in the direction they crossed the line. “3D” and “stereo” mean the same thing in this industry, and since about 2010 stereo has been the technology to beat for retail; nearly every enterprise vendor, ourselves included, builds on it.

Variants

Standard units cover ceilings up to about 6 m (3D Scope II LC coverage chart); high-mount models rated to 20–25 m cover atriums and concourses.

Verdict: the most complete counter for a bright, busy retail entrance — groups, children, carts and U-turns handled by measurement — at the price of a cable run and enough light.

3D Scope II LC: stereo with on-device AI and HDR, 99% rated with a 95% floor guaranteed in writing for two years, one PoE cable under 8 W, coverage up to 8 × 8 m per sensor (8 × 22.8 m with two), in/out, adults and children, carts, groups, wheelchairs, U-turns, staff exclusion, video audit on request; $1,149.95. Where we lose: ceilings above 6 m, and age and gender, which the 3D Scope II LC does not estimate.

A time-of-flight sensor emits pulses of infrared light and times their return from every point below; the round-trip time gives distance, and distance from the ceiling gives height. The result is a depth map very like a stereo sensor’s — heads, carts and children distinguished by height, groups separated, direction known — but produced without any color image at all. Popularized by the Microsoft Kinect, ToF was “new and promising” when we last wrote about it in 2016. It has since matured into a real alternative to stereo from several specialist vendors.

Variants

LiDAR units apply the same principle with scanning lasers over larger areas.

Verdict: the choice when a rule says “no cameras” and you still want depth-class counting. Test it in daylight at your own entrance, and expect to need more units than stereo on a wide one.

AI in people counting comes in two kinds, and not every AI counter processes on the device. Cloud AI sends images or live video to a server or the cloud for processing. Edge AI processes on the sensor itself, so only metadata leaves it — ins and outs, adult and child counts, dwell time and heat-map data. Every SMS Storetraffic sensor that uses AI is edge AI: no images or video are sent to a server or the cloud for processing. The difference between the two deserves its own article.

“Edge AI” on a 2026 datasheet means the sensor runs a neural network on the device to classify what it sees, rather than applying hand-written rules. On a stereo or ToF sensor the network works on depth data: it learns what a head, a cart, a wheelchair, a stroller and a child look like from above, so the sensor can count carts separately, count a wheelchair user as a person, tell a tall child from a short adult better than a fixed height threshold can, and separate a tight group. On thermal sensors, machine learning has closed much of the accuracy gap to cameras. On 2D video, AI object detection rescued a technology that background subtraction had left behind.

Three things it does not mean. It does not mean video is being sent to the cloud — the whole point of edge processing is that only metadata leaves the device. It does not mean face recognition; a counter trained on head shapes from above never sees a face. And it does not mean accuracy is solved: an AI sensor mounted over a door swing, in the dark, or above its rated height still counts badly. The useful questions are the same as ever — what does the sensor capture, what leaves it, and what is the guaranteed floor.

Where AI adds capabilities that need more care: age and gender estimation, and re-identification — recognizing the same person again, across sensors or on a later visit. How each is built decides its legal footing. Age and gender estimated on the sensor, with only statistics leaving it and no one identified, is not treated by EU regulators as biometric identification; Canada’s privacy commissioners found that face-based age and gender estimation in shopping malls required express consent, and Illinois’ biometric law covers any scan of face geometry. Re-identification goes further, because its purpose is to single one person out: some systems do it with face recognition, and some send images of faces to a server to do it. For either feature, ask what it works from, where the processing happens and what leaves the sensor before you switch it on. Retailers that got this wrong made the news for tracking their customers. In the US, a pharmacy chain was banned from facial recognition for five years, and cameras on pharmacy and grocery shelves that could guess shoppers’ age and gender made national news. In Canada, a mall operator collected five million shoppers’ images to estimate their age and gender, and a dozen hardware stores were found to have broken privacy law with facial recognition. The damage to their reputation was far larger than anything the data was worth.

Sensors that listen for wireless signals — Wi-Fi, Bluetooth or cellular — count devices, not people, and not only phones. That is the whole limitation, and no amount of signal processing removes it.

“Phone counting” covers three different technologies, and they measure different things:

A device is not a person, and phones are not the only devices that broadcast. Wi-Fi and Bluetooth sensing also pick up smartwatches, fitness trackers, wireless earbuds, tablets, laptops, e-readers, handheld game consoles, Bluetooth item trackers, portable hotspots and the car waiting at the curb; cellular sensing picks up anything with a mobile radio, from cellular watches and tablets to connected cars and card terminals on a mobile connection. One shopper with a phone, a watch and earbuds can count as several devices; a child with none counts zero. A device that is asleep, has Wi-Fi off or is not transmitting is not counted at all, while fixed devices in and around the store transmit all day and have to be filtered out. And these signals pass through walls, so the sensors cannot cleanly separate shoppers inside your store from people walking past it.

Wi-Fi and Bluetooth counting became much harder when Apple and Google began disguising phones’ hardware addresses from 2014–2015. Researchers can still estimate how many broadcasting phones are nearby — the best recent method reports about 92% accuracy — but that is 92% of broadcasting phones in range of a receiver, not 92% of people, and not people crossing a door. To turn phones into people, the researchers had to calibrate against a separate people count.

Cellular counting measures how much mobile-phone radio energy is in the air and estimates from it how many phones are nearby. It identifies no one, which is to its credit, but it reflects what phones are doing rather than how many people are present. Reading a phone’s actual identity from its signal is off limits to any commercial counter: in the United States under the federal Pen Register Act and section 605 of the Communications Act, in Canada under section 9 of the Radiocommunication Act.

All of these methods also depend on how phones behave today, which phone makers and standards bodies keep changing to protect privacy. A method that works this year can stop working after a software update.

Phone-signal data is useful for dwell time, zone flow and trade-area context in a large venue. It is not a door count.

The radio technology that is useful at a door in 2026 is ultra-wideband, in the other direction: staff carry a UWB tag and a receiver on the sensor tells the counter to ignore them (how our staff exclusion works). Consent is built in, because the only devices tracked are the ones you handed out, and the tags identify no individual.

Every serious people counter now counts in both directions. Overhead sensors track each person across their view and know which way they crossed the line; directional beams such as PEARL use two beams a few centimeters apart, so A then B is an entry and B then A an exit. IN is the number almost everything is built on: traffic, conversion rate and staffing all use the count of people coming in, because the moment someone walks in is the moment they may need help to become a customer.

It was not always this way. Many early counters, single beams above all, could not tell direction, so the industry took the total count and divided it by two: over a day the people who come in are the people who go out, so half the total lands close to the IN count. With directional counting standard, there is no need to estimate a figure the sensor simply reports.

OUT earns its place when you need to know what happens after the door: occupancy, time spent in the store, and how customers flow between entrances. Occupancy is the one to handle with care. It is cumulative IN minus cumulative OUT, so every counting error persists. A 95%-accurate sensor at a door seeing 100 people an hour makes about five errors an hour; if they lean one way, and they do, the system believes there are 25 phantom people in the store by closing and 60 by Sunday night. That is arithmetic, not a fault, and it matters for staffing: an accumulated error makes the store look fuller than it is, which can lead you to staff more than necessary. Live occupancy is a legitimate use — for capacity limits, gyms, study halls, event floors — but it needs a high-accuracy sensor on every entrance, a daily reset, and a manual correction button, which is how our Real-Time plan is built.

The “group counting” feature

Some sensors offer to count people who enter within a set distance of each other as one “shopping unit”, which raises the reported conversion rate. The distance rule cannot know whether two people are a couple, colleagues or strangers who arrived together, so it is arbitrary at best and flattering by design. We have offered it and we advise against it: count people, know that some of them shop in groups, and measure your improvement against a consistent baseline.

Our own specifications come from our product pages and help center, linked in place. Industry accuracy ranges are the figures vendors publish for their own products, summarized without naming them; no competitor is quoted or linked on this page. Named comparisons live on the comparison page.

How this guide was made

This page replaces the three-part “How customer counter technology works” series we published in 2016, keeping what has stayed true — the 54-inch rule, the beam’s failure list, the thermal standing-still problem, the occupancy drift arithmetic, the skepticism about group counting and Wi-Fi — and updating what changed: time-of-flight matured, edge AI arrived, Wi-Fi counting ended, and prices fell. SMS Storetraffic sells two of the technologies described; we say so where they appear and name other vendors where they fit better. Corrections via our contact page.