Without camera and without tag: radar, ToF, LiDAR, heat

All technologies from previous articles have a common border. The camera is not placed in the locker room, shower and steam room — this is unacceptable. The bracelet does not answer the question of what the guest is doing in the zone, and no one applies it with wet hands. The radio tag is silent under the water, but in the locker room it lies in the locker with the phone.

At the same time, it is these zones that are the most problematic at the facility. Wet tiles, temperature changes, elderly guests, alcohol, cramped space with lockers. A fall in the shower happens incomparably more often than drowning in a bowl, and a guest who falls asleep in the steam room is a real, not a theoretical risk.

This article is about a class of sensors that closes this particular hole: they determine what a person is, how many of them, whether he is standing or lying down, and at the same time they do not form an image in principle. Not a "camera with a smudged face", but a device that does not produce an image at all.

What is this class

General principle: the sensor measures a physical quantity, from which the presence of a person follows, but from which it is impossible to reconstruct what he looks like.

Type	What measures	What he can do	The image of a person
Radar 60 GHz	Reflection of radio waves, Doppler shift	Presence, counting, posture, Falling, breathing	Point cloud, no image
Flight time sensor	Distance to the surface by light return time	Account in the opening, occupancy, height	There is no depth, color, or face map
LiDAR	The same, scanning in the sector	Account and position in the volume of the premises	Point cloud
3D stereo	Depth on two lenses	The exact score in the opening	Contains cameras, processing on the device
The thermal matrix	Infrared radiation of the body	Presence, counting, rough posture	Heat spots
Passive IR (PIR)	Changing the thermal field	Is there movement or not	I don't
Floor vibration	Mechanical vibrations	Steps, fall	I don't

Two remarks about this table.

3D stereo stands alone. This is the most accurate way to count people in the opening, and it is based on two lenses, that is, there are cameras inside the device. Manufacturers emphasize that processing is performed on the sensor itself and only numbers go out — for example, Xovis describes just such a scheme and claims a counting accuracy of up to 99%. For the legal circuit, the difference between "the camera is inside, but the image does not leave the device and is not recorded" and "there is no camera at all" is significant, and for sensitive areas it is decided in favor of radar.

Passive IR is underestimated. The cheapest sensor of all, it does not count people and does not determine the pose, but it honestly answers "there is someone in the booth" or "there is movement in the steam room." For some of the tasks, this is enough, and sometimes it's worth starting with it.

Millimeter-wave radar: the main tool for sensitive areas

How it works. The sensor emits radio waves in the range of about 60 GHz and analyzes reflections: the distance is determined by the delay, the speed of movement is determined by the frequency shift. The result is a set of points with coordinates and velocities, from which the algorithm collects "targets" and classifies their behavior.

What he can do. Presence (including a stationary person — due to micro—movements of breathing), counting people in the room, position with an accuracy of about tens of centimeters - meters, height of the target, which means the difference between "standing" from "sitting" and "lying", and a sharp change in position, that is, falling.

Silicon and suppliers. A common basis is the Texas Instruments transceivers of the IWR6843 family, including the version with an antenna in the case; such solutions are also produced by Infineon, Vayyar and a number of manufacturers of ready—made modules. In the scientific work of 2026 devoted to counting people and classifying movement at 60 GHz, the authors directly formulate the motivation: traditional methods on cameras and infrared sensors carry risks to privacy, depend on lighting and do not work well in a difficult environment.

Fall detection. This is an actively developing area, and it is the work about wet rooms that is important for us. In 2026, the architecture of fall detection in the bathroom was published, combining a radar and a floor vibration sensor, tested on a full-scale mock-up of the bathroom. There are also works combining radar with three-dimensional measurement of vibrations for the sake of privacy, and works on fall detection, resistant to changes in the situation. Separately — the scheme of private monitoring of the bathroom on passive IR sensors and LiDAR.

The direction of thought here is the same: one sensor is not reliable enough, the combination of two physical principles — radar plus vibration, IR plus rangefinder — gives noticeably fewer false positives. This is worth remembering when a supplier offers "one sensor per room".

About vendor numbers. In commercial materials, there is "99% drop detection accuracy without cameras and wearable devices." These are statements without a description of the conditions; in scientific papers, the same class of problems is formulated as unsolved, indicating insufficient stability in different rooms. A reasonable position: presence detection and counting are mature functions, fall detection is a function that needs to be checked specifically in your room, with simulated events.

Radar limitations.

• Metal and mirrors. The locker room with steel lockers and mirrors is a reflective camera in the radio—frequency sense. This gives phantom goals, and this is the main reason why the arrangement needs to be checked on the spot.
• Glass and partitions. The radar partially passes through thin non—metallic partitions - this is both a virtue (it sees behind the screen) and a problem (it sees the next room). The zone boundaries are configured programmatically, and the setting must be checked.
• Sector and range. The sensor sees a cone, not a room. Planning is like with cameras: by zones, with overlap.
• Setting up for a room. Behavior classification algorithms are sensitive to geometry. Transferring settings from another object usually requires fine-tuning.
• Operating conditions. About this below separately, because for the steam room this is a crucial issue.

Time-of-flight sensors, 3D stereo and LiDAR: counting

For entrances and openings, this is the main tool — more precisely, video analytics, cheaper and without legal complications.

Declared accuracy. Ranges for open reviews: flight time sensors — 93-97%, LiDAR — 95-99%, 3D stereo — 95-99%. Manufacturers claim an upper limit: Xovis — up to 99%, Terabee — typically 98% and higher in most conditions. All these figures are vendor and review, independent tests could not be found in the public domain.

The main enemy is daylight. Terabee himself notes that external lighting affects the infrared signal; in tone-independent reviews, there is an observation that inexpensive time-of-flight sensors at entrances with real daylight are noticeably more mistaken than in the specification. For the entrance group with panoramic glazing — and this is a typical thermal architecture — the question of sensor behavior in direct sunlight should be asked directly and checked.

The second enemy is a dense slow stream. People walking close and stopping in the doorway is the worst scenario for any meter.

LiDAR does not cover the opening, but the volume: according to survey data, the radius is about 5-10 meters, in terms of solutions up to 25 are claimed. It works confidently in the dark and in difficult lighting, costs more and is used where it is necessary to count and roughly position people in the hall, and not in the door.

Arithmetic that spoils all employment projects

This is the most practical section of the article, and it is almost never talked about at the sales stage.

The counter on the door counts the entrances and exits. Current employment is the difference in accumulated amounts. And the error also accumulates.

A simple example from a review of counting people: a sensor with 95% accuracy on a door through which one hundred people pass per hour. By the end of the day, the system can show twenty-five people in the building who are not there.

Hence the three mandatory requirements for any download control project:

1. Daily reset at the moment when the object is guaranteed to be empty, plus a manual correction procedure.
2. Reconciliation with an independent source. The turnstile with bracelets gives a list of those who entered and those who did not leave — this is a different physics and a different type of error, and the discrepancy between the two sources shows the quality of both.
3. Understanding which metric you need. The flow over the period (how many people have passed) is measured reliably. Instant occupancy is a metric with an accumulating error, and if you are building capacity control on it, you need both of the previous points.

Our facilities: steam room, shower, swimming pool, changing room

The main thing here is what you need to talk about with the supplier, and the only thing that cannot be learned from the reviews.

Steam room and sauna. 90-110 °C, humidity up to 100%, in the hammam — condensation on all surfaces. Typical presence sensors are designed for ordinary rooms, and their operating temperature range ends far from the conditions of the steam room. Manufacturers do not publish this parameter in marketing materials. The requirement is simple: request in writing the operating range of temperature and humidity, resistance to condensation and degree of protection according to IP, and do not accept the answer "we put it in a similar place." Practical schemes for a steam room are usually one of two: a sensor in the dressing room that controls the entrance and exit, or a sensor in the steam room itself in a special design — and the second option almost always requires individual selection.

Shower room. High humidity, splashes, steam, but the temperature is normal. The task is doable, moisture performance is required and a layout that takes into account the cabins.

Pool and bowl area. Splashes, chlorinated atmosphere (aggressive to electronics and metal), glare are essential for optical sensors, not for radar. Chlorine in the air is a separate factor of durability, and the service life in such conditions should be asked separately.

The locker room. Metal lockers, mirrors, narrow aisles, benches. For radar— a difficult reflective environment; for time—of-flight sensors from above, it is quite working. Typical scheme: presence sensor in sections plus fall detection as a separate function.

Scenarios by zone

Zone	Task	Than closes	Why is that so
Locker room	Presence, fall, employment of the section	60 GHz radar, time-of-flight sensor from above	Camera is not allowed, the label is in the locker
Shower room	Cabin occupancy, falling	Radar, passive IR for employment	Camera is not allowed
Steam room, sauna	Is there a person lying, is the time exceeded	The sensor is made according to the conditions plus the mark of the bracelet at the entrance	The camera is unacceptable, the optics go blind in pairs
Toilet	Employment of cabins, cleaning after the fact	Passive IR, time-of-flight sensor	The simplest and most cost-effective automation
Entrance group	Flow, throughput	3D stereo or time-of-flight sensor	More precisely, video analytics, without legal complications
Zone opening	Zone loading	Flight time sensor	Cheaper than the coordinate system
Recreation area	Employment of sun beds and seats	LiDAR or radar	There are no labels, you need a volume, not an opening
Hazardous area at work	Human presence	Radar	Works in the dust and in the dark
Ward, medical room	Presence, fall	Radar	Patient privacy

Retail, shopping center, HoReCa

For these objects, the counter at the entrance is not an auxiliary tool, but the foundation of all analytics: without a denominator (how many people entered), any metric turns into an absolute number without meaning. Normalization to the input stream is what distinguishes a meaningful heat map from a beautiful picture, and this is discussed in detail in the article about metrics.

A reasonable architecture for a retail facility: 3D counters at the entrances, time-of-flight sensors at the entrances to anchor zones, LiDAR or radar in large open spaces, video analytics only where the context is needed (cash register, conflict situations). It is cheaper, more stable and legally calmer than any attempt to get the same thing via Wi-Fi or cameras.

Industry and medicine

Presence in the danger zone. Radar sees a person in dust, fog, darkness and behind a thin barrier — that is, where the camera and optics do not work. Typical application: blocking equipment when a person enters the work area.

Single shift. A person works alone; you need to know that he is conscious and moving. A radar with immobility detection is a working circuit that complements the button (you may not be able to press the button).

Ward and medical room. Patient privacy makes the camera unacceptable, and falling is the main risk. This is the most mature niche of radar detection of falls, and the experience is transferred from there to objects with guests.

What is the law in Russia

The main advantage of this class of sensors is legal.

The number of people in the room and the point cloud, from which it is impossible to restore the appearance of a person, as a rule, do not relate to personal data. No image, no biometrics, no ID. This is a fundamentally quieter mode than video surveillance, and that is why such sensors are a legitimate way to monitor areas where cameras are unacceptable.

But there is a border, and it is worth knowing.

One person in the room. If there is one guest or one employee in the room at the moment, the "impersonal" message about his presence and position actually refers to a specific person. In conjunction with the shift schedule or with the mark of the bracelet at the entrance, this is already information about a specific person.

A bundle with other systems. The presence sensor in the steam room itself is impersonal; the same sensor, time-matched with the event "bracelet No. 1428 entered the zone", gives personal data. This is not prohibited — it requires proper registration.

The staff. Monitoring the presence of an employee in the zone is a matter of labor legislation and local acts, regardless of whether it is done by a camera or radar.

Information is still needed. Legal admissibility does not negate ethics and reputation. A guest who finds out that there was "some kind of sensor" in the locker room, about which he was not warned, will take it as surveillance, even if everything is formally legal. The correct practice is a short and clear wording in the rules of the object and on the information plate: what is measured, for what, that images are not formed. The analysis of the legal outline is in a separate article in the series.

Restrictions and how they cost

There is no identity. The sensor knows that the person is, but does not know who he is. Bypass: the mark of the bracelet at the entrance to the zone gives a binding, the sensor inside — the fact of presence and status.

Accumulation of employment error. Bypass: daily reset, checking with the turnstile, using the flow instead of instant employment where possible.

Sector and range. Bypass: sensor density, overlap, zone plan, not "one sensor per room".

Phantoms of metal and mirrors. Bypass: on-site inspection, adjustment of zone boundaries, if necessary, a combination of two physical principles.

Steam room conditions. Bypass: a sensor in the dressing room instead of a steam room, individual selection of execution, time control on the bracelet as the main mechanism and a sensor as an additional one.

Immaturity of fall detection. Bypass: tests with simulation in your room, a combination of radar with a different principle, a reasonable threshold and a staff inspection procedure.

Checklist for talking to the supplier

1. Does the device form an image, and if there are optics inside — what exactly leaves the device?
2. Operating temperature and humidity range, resistance to condensation, degree of protection — in writing.
3. What is the service life declared in a chlorinated and humid atmosphere?
4. What sector and what range, how many sensors do we need for our plan?
5. How does the sensor behave next to metal lockers and mirrors?
6. How are the zone boundaries set up and does the adjacent room get in?
7. For meters: what is the accuracy in direct sunlight and in dense slow flow?
8. How is the accumulation of a busy error solved: reset, correction, reconciliation?
9. For fall detection: what data is the model trained on, are there implementations in wet areas, is it possible to conduct tests with simulation?
10. What is transmitted to our system: numbers, events, a point cloud? Is there an API?
11. How do we know if the sensor has failed?
12. What happens when the connection is broken: are events buffered or lost?

Sources

Sources

• Counting people and classifying movement on radar 60 GHz, 2026
• Fall detection in the bathroom: radar and floor vibration, 2026
• Private multimodal fall detection system for bathrooms
• Fall detection on radar, resistant to changes in the situation
• Private monitoring of the bathroom on IR sensors and LiDAR
• Xovis: 3D counting sensor - claimed accuracy and processing on the device (manufacturer's data)
• Terabee: measurement of counting accuracy — declared accuracy and influence of external light (manufacturer's data)
• Comparison of methods of counting people, 2026 — accumulation of employment error
• Overview of counting technologies: beam, heat, stereo, flight time, Wi-Fi — accuracy ranges (supplier overview)