People Counting Sensor: ToF vs Stereo Vision, Which Is More Accurate?

You want to count the people walking through your store entrance or lobby, and you are looking at two near-identical devices: one people counting sensor based on time-of-flight and one with two camera lenses. Here is the short answer: under good conditions both methods deliver very similar results. Milesight quotes up to 99.8 percent accuracy for its VS13x ToF series and for its VS125 stereo counter alike. Which one counts more accurately at your entrance almost always comes down to mounting height, coverage and lighting, not the sensing principle on its own.

So understand both technologies, then take a tape measure to your doorway. Using the datasheet figures for the Milesight VS135 (ToF) and the Milesight VS125-P (stereo), you can work out before you buy whether a device will even cover your entrance.

How does a ToF people counter work?

A time-of-flight sensor brings its own light. In the VS135, that is invisible infrared at a wavelength of 940 nm. The light hits heads, shoulders and the floor and bounces back. For every pixel, the sensor measures how long the light took to travel there and back and converts that into a distance. The result is a depth map, not a photo: a height profile of the scene below.

In that depth map, people stand out as clear bumps above the floor. An AI model recognizes those shapes, assigns each one a track and follows it across several measurements. When a track crosses the configured counting line, the device logs an entry or an exit depending on the direction. Because distance is measured directly, the counter knows each person’s height. Milesight specifies a distance measuring accuracy of ±3.5 cm for the standard model and ±6.5 cm for the high-ceiling version.

A 3D people counter like this needs no ambient light, since it supplies its own illumination, which can be set to automatic or always on. Its limit is the measuring range: the standard VS135 measures distances from 0.5 to 3.5 m, the high-ceiling version from 2 to 6.5 m. Outside that window, the time-of-flight sensor returns no usable data.

How does a stereo vision people counter work?

A stereo counter works like your two eyes. Two image sensors sit a fixed distance apart; in the VS125-P, they are two 4-megapixel sensors. Any point in the scene shows up at a slightly shifted position in each image, and the closer it is to the device, the bigger that shift. From the shift, the counter calculates distance by triangulation and also gets a three-dimensional view.

On top of that, a deep learning model analyzes the camera images themselves. That is the key difference from ToF: a stereo vision counter sees outlines and texture as well as depth, which lets it derive extra attributes such as groups or the difference between adults and children. For the depth calculation to work, though, the software has to find the same points in both images. That works best on surfaces with some structure and contrast.

For darkness, the VS125-P has built-in infrared illuminators, and Milesight lists the required illumination as 0 lux with the IR illuminators on. Its field of view is 101° horizontal and 70° vertical, and the supported mounting height runs from 2.2 to 6 m. One model therefore covers both ordinary entrances and taller halls.

ToF vs stereo people counting: which is more accurate?

On paper, there is no winner. Both manufacturer figures reach up to 99.8 percent, and those numbers come from test conditions: correct height, clear view, a well-placed counting line. At a real entrance, both devices lose accuracy for the same reasons. The counting line sits too close to the door edge, a door leaf swings through the view, a display stand blocks one side, or the unit hangs slightly crooked after someone worked on the ceiling.

Each method has its own strengths. ToF measures depth directly, regardless of whether the floor is patterned or plain, which keeps head detection stable in crowds and with people in similar clothing. Stereo calculates depth but works with high-resolution images and a wide field of view that captures more area at greater heights. For dense crowds, Milesight also lists a dedicated occlusion detection feature for the VS135.

The only way to know whether your counter is accurate is a cross-check. Count by hand for half an hour at peak time and compare each direction separately. If people keep going missing along the same edge, that points to tight coverage, not weak technology.

Which people counting sensor copes better with sunlight and darkness?

Backlight is the classic problem at glass entrances. Low winter sun shines through the door, casts hard shadows across the floor and creates bright patches on polished stone. For a camera, those are strong contrasts that shift every time someone walks past, and a stereo counter can misread moving shadows more easily because it compares image content.

A ToF sensor cares little about shadows because it measures distance with its own light. It is not immune, though. Direct sunlight also contains infrared and can add noise to the signal. The 940 nm wavelength is used partly because sunlight is relatively weak in that band. Mirrors, glass panels and very glossy floors can redirect the emitted light and distort individual distance readings.

In complete darkness, both devices keep working: the VS135 with its active ToF illumination, the VS125-P with its IR illuminators. Neither manufacturer gives a sunlight limit in lux. So place the counting zone one to two meters inside the glass front, where the sun patch does not reach, and test the counter at exactly the time of day when the sun is at its worst angle.

What mounting height does a people counter need?

An overhead people counter hangs facing straight down above the passage. Measure the mounting height from the finished floor to the underside of the device, not the room height from the floor plan. A suspended ceiling, a beam or a bracket can easily change the figure by 20 or 30 cm, and that is enough to fall outside the supported range.

The device limits are clear: the standard VS135 is designed for mounting heights up to 3.5 m, the VS125-P for 2.2 to 6 m. Above a typical door with about 2.10 m of clearance, a counter usually ends up between 2.5 and 3.5 m, within range for both devices. For lower situations, both models offer lintel mounting and extended ceiling mounting with an optional bracket in addition to standard ceiling mounting.

At a mounting height of 2.5 m, only about 80 cm remain between the device and head height. The field of view there is correspondingly narrow, and the counter sees each person over only a short stretch of their path. The shorter that stretch, the fewer measurements the software has to determine direction reliably.

Which people counter works on a high ceiling?

As soon as the device has to hang higher than 3.5 m, the standard VS135 is out. Typical cases are entrance halls, atriums and open sales floors. You then have three options: the VS135 high-ceiling version for mounting heights up to 6.5 m, the VS125-P up to 6 m, or an extension mount that brings the standard counter down to a suitable height.

The high-ceiling version is not a standard unit with longer reach; it has different optics. Its field of view of 60° horizontal and 45° vertical is much narrower than the 98° by 80° of the standard model, and its measuring range only starts at 2 m. Two practical rules follow. First, the high-ceiling version does not belong just above a normal door, because people’s heads would be too close to the sensor. Second, the extra height does not automatically give it wider coverage than a standard unit at 3.5 m.

Plan the power supply as well. Both VS135 versions need PoE+ to 802.3at or a 12 V adapter, and the high-ceiling version draws 10 W on average and 24 W at peak. Your switch has to deliver that per port, and at 6 m you should also plan how you will reach the device for maintenance later.

How wide an entrance can one people counting sensor cover?

What matters is not the area on the floor but the area at head height, where the counter has to detect and track people. If you calculate with a head height of about 1.7 m and the datasheet field of view, you arrive at exactly the figures Milesight publishes: 10.43 × 6.02 m for the VS125 at 6 m and 22.05 m² for the VS135 high-ceiling version at 6.5 m. The same calculation gives this overview for a single device, width by depth at head height, rounded:

The table shows two things that surprise many people. Below a 3 m mounting height, no device covers more than about two meters of width. And at great heights, the VS125-P captures almost twice the width of the VS135 high-ceiling version. Mount the counter so the wide side of its field of view runs across the entrance and the short side points in the walking direction. The values describe the field of view; the reliably usable zone is somewhat smaller toward the edges. Leave a buffer on each side and check your result against the manufacturer’s coverage calculator.

For wide entrances, such as a 5 m double sliding door or an open mall portal, one device is often not enough. In that case, you link several counters into one shared counting area by stitching. Milesight lists up to 16 devices for the VS125-P, and the VS135 also supports stitching multiple units. The fields of view should overlap slightly, and that transition zone deserves extra attention during your cross-check.

Can a people counting sensor tell children, groups and carts apart?

Yes, but not every model does everything, and not at every height. According to its datasheet, the VS135 supports group counting, adult and child differentiation, staff detection and even shopping cart fill level detection. It also has U-turn filtering for people who step into the zone briefly and turn back, and tailgating detection for someone passing closely behind another person. Because ToF measures body height directly, telling children from adults rests on a solid basis here.

The VS125-P also handles group counting, U-turn filtering, adult and child differentiation and staff detection. Milesight, however, specifies the last two only for mounting heights of 2.2 to 4 m. If the stereo counter hangs in a 6 m hall, it still counts reliably but no longer distinguishes children. Milesight does not list shopping cart detection for the VS125-P.

A common setup mistake: a child carried or in a stroller may be treated differently from a child walking alone. Decide in advance what counts as a visitor in your reports, and test exactly those situations during your cross-check.

Milesight VS135 vs VS125-P: which people counter fits your entrance?

Both devices count in both directions across up to four counting lines, analyze up to four areas with dwell time, store up to one million records locally and send their data over HTTP(S) or MQTT(S). The differences lie in height, field of view, features and power:

The VS135 suits you if your retail project needs more than simple ins and outs, such as shopping carts, tailgating or children, and if your entrance has strong shadows or changing light. Choose the standard version for normal ceiling heights and the high-ceiling version above 3.5 m. The VS125-P is the more flexible choice if you want one device for very different heights or need to cover as much width as possible with a single counter in a tall hall. It also runs from an ordinary PoE port.

Where can you buy a people counting sensor?

Before you decide, measure three things at your entrance: the actual mounting height from the floor to the underside of the device, the walkable width including any side doors that open, and the clear path in front of and behind the door. The height tells you which models qualify at all: the standard VS135 up to 3.5 m, the VS125-P from 2.2 to 6 m, the VS135 high-ceiling version up to 6.5 m. The width and the coverage table show whether one device is enough or whether you need to stitch several together. Then look at the light: glass fronts with low sun favor ToF. Think about which extra features you will actually use, whether children, groups or shopping carts, and check whether your switch delivers PoE or PoE+. Finally, schedule a peak-time cross-check before you trust the numbers. That way, your people counting sensor will be as accurate at your entrance as the technology allows. If you are looking for a suitable counter, you will find our models in the people counter category.