Accurate Visitor Counting System People Counting Accuracy, High Accuracy People Counters, and What a 98 Percent Claim Actually Covers
Every people counting vendor that publishes a number publishes something between 95 and 99.9 percent, and those numbers are not comparable, because they are not measuring the same thing. Almost all of them describe crossings at a single well-lit doorway under moderate flow. None of them describe the number most buyers actually care about, which is how wrong the occupancy figure on the dashboard is at 4pm on a Saturday. Those are different quantities, and the gap between them is the whole subject of this guide.
How Accurate Is People Counting?
Overhead 3D stereo sensors are the accuracy leader at above 98 percent on a controlled doorway. Break-beam and infrared counters run 85 to 95 percent, thermal counters around 85 percent, and software counting from an existing security camera typically 70 to 90 percent depending almost entirely on the mounting angle. Those figures describe individual crossings. Occupancy, which is entries minus exits, is a much harder number and degrades faster than any of them.
A practical rule that survives contact with real buildings: retailers hold a counter to 95 percent or better against a manual count before they will trust conversion numbers derived from it. That is a sensible bar, and it is a bar you set yourself with a clicker, not one you accept from a datasheet.
High Accuracy People Counting, Ranked by What Actually Limits Each Technology
Accuracy tables usually stop at a percentage. The more useful column is the failure mode, because that is what tells you whether the technology will hold up in your specific doorway. A counter that fails on groups is fine in an office lobby and useless at a cinema when a screening lets out.
| Technology | Typical accuracy | What actually breaks it | Best fit |
|---|---|---|---|
| Overhead 3D stereo | Above 98 percent | Very little, which is why it costs what it costs. Height requirements and doorway width are the practical constraints, not the algorithm. | Main entrances where the number carries money |
| Break-beam and infrared | 85 to 95 percent | Groups. Two people abreast break one beam once. Also blind to direction unless you pair beams. | Quiet single-file doorways, cheap coverage of side doors |
| Thermal | Around 85 percent | Ambient heat. Sun through a glass storefront and heaters near the door both degrade it, which is exactly where doors are. | Privacy-sensitive sites where no image may be captured |
| Dedicated AI camera counter | Typically 95 percent and up | Crowd density and occlusion. Accuracy falls as people per minute rises, on every camera system without exception. | Moderate-traffic doorways where you also want dwell data |
| Software on an existing security camera | 70 to 90 percent, higher on a downward-facing door camera | The mounting angle, overwhelmingly. A camera hung at head height for face capture is looking at a wall of overlapping bodies. | Multi-door and multi-site trend data at low marginal cost |
The pattern across the whole table is that geometry beats software. An overhead view of a controlled doorway will out-count a more sophisticated algorithm looking across a crowded, sunlit atrium, every time. Before you shortlist vendors on accuracy claims, go and look at where your cameras actually point. The full hardware cost picture is in our companion guide to people counting sensor costs by technology.
A 98 Percent Accurate Counter Does Not Give You a 98 Percent Accurate Occupancy Number
This is the single most expensive misunderstanding in the category, and it is pure arithmetic rather than opinion. Footfall and occupancy are computed differently, so identical counting errors behave completely differently in each.
Footfall is cumulative. Count 2,000 entries at 98 percent accuracy and you report roughly 1,960 instead of 2,000. You are 2 percent light, and tomorrow you will be about 2 percent light again, which means your week-on-week comparison is almost perfect even though the absolute number is wrong. Consistent bias is harmless to trend analysis. That is why footfall reporting tolerates imperfect counters so well.
Occupancy is a difference, not a total. It is entries minus exits, accumulated since the last reset. Take the same doorway: 2,000 entries and 2,000 exits over a day at 98 percent per-event accuracy means roughly 40 missed entries and roughly 40 missed exits. If those errors happened to cancel, the occupancy figure is perfect. They do not cancel on demand. The two error streams are independent, so the running difference wanders, and by close of business the occupancy figure can be off by tens of people while the footfall total is still only 2 percent light.
Two structural consequences follow. First, occupancy error does not average out over the day, it accumulates, and only a genuine empty-building reset clears it. Second, a single uncounted door is catastrophic for occupancy and merely annoying for footfall. If staff use a back door 40 times a day and that door has no counter, your footfall number is unaffected and your occupancy number drifts by 40 people every single day, in one direction, forever.
What to do with this: if you are buying for footfall and conversion, per-crossing accuracy is the right spec and a 95 percent counter is genuinely fine. If you are buying for live occupancy, ask two different questions instead: is every route in and out counted, and does the system reset to zero when the building is verifiably empty. Vendors answer the first question with a percentage. Only the second pair of questions predicts whether the dashboard will be right at 4pm.
Which Vendors Publish an Accuracy Figure, and What Their Wording Commits Them To
We read the public sites of the best known people counting vendors in August 2026 and recorded the exact accuracy language. The wording matters more than the number. There is a real difference between a guarantee, an up-to figure and silence, and buyers routinely treat all three as though they were the same claim.
| Vendor | Published accuracy language | How strong that wording is |
|---|---|---|
| SenSource | Guarantees 98 percent accuracy, and notes many customers see above 99 percent, backed by fleet monitoring | Strongest. A guarantee with monitoring behind it is a commitment, not a marketing figure. It also publishes guidance on comparing accuracy claims, which is unusual candor. |
| V-Count | Up to 99 percent bi-directional accuracy | Medium. Up-to phrasing describes a ceiling under favorable conditions rather than a floor you can hold them to. Note it does specify bi-directional, which is the harder case and worth credit. |
| Xovis | No accuracy percentage published | No public commitment. Strong reputation in airports and transit, but you will be relying on references and a pilot rather than a datasheet number. |
| FootfallCam | No accuracy percentage on the homepage | No public commitment at the top level. Pricing and specification both run through local distributors, so ask for the accuracy claim in writing with the quote. |
| Density | No accuracy figure on its pricing page | Positioned for workplace occupancy in rooms and at desks rather than doorway counting, so a doorway accuracy number would not be the relevant spec anyway. |
| RetailNext | Nothing published | No public accuracy or pricing information at all. Everything happens inside a sales process. |
We do not publish a single accuracy percentage for our own counting, and the reason is in the table above: any number we printed would describe our test doorway rather than yours, where the camera angle dominates the outcome. We would rather hand you the measurement method below and have you produce a figure that is actually about your building. If you want the full vendor-by-vendor comparison including pricing, that lives in our roundup of the best people counting software compared for 2026.
How to Measure Counting Accuracy in Your Own Building
Pick two hours, not one
One busy hour and one quiet hour. Accuracy falls with density, so a single quiet-hour test tells you almost nothing about the number you will see on a Saturday.
Count in and out separately
Two clickers, or two people. A system can be excellent on entries and poor on exits, and a combined tally hides exactly the asymmetry that ruins occupancy.
Log the staff separately too
Employees are a large share of crossings at a small site. If staff exclusion is not configured, you are measuring the counter against a ground truth it was never meant to match.
Report error with a sign
Under-counting by 4 percent and over-counting by 4 percent are both 96 percent accurate and they behave completely differently in an occupancy figure. Direction of bias is the useful output.
Run the empty-building check
At close, when the building is verifiably empty, read the occupancy figure. Whatever it says instead of zero is your daily drift. This one number is worth more than any datasheet.
Repeat after any change
A moved display, a new door mat, a seasonal window vinyl or a camera someone nudged on a ladder can all move accuracy. Re-validate whenever the doorway changes.
That empty-building check in step five is the cheapest audit in retail operations and almost nobody runs it. It takes ten seconds at close and it is the only measurement that catches an uncounted door, which is the failure that quietly poisons every occupancy-derived decision downstream.
How Much Accuracy You Actually Need to Buy
Accuracy is not free and past a point it stops changing decisions. The question to settle before you shortlist anything is what the count is for, because the answer sets the bar and the bar sets the budget.
If footfall feeds a rent calculation, a landlord report or paid audience reporting, you are in the money-changes-hands category and you should buy overhead 3D stereo at the doors that matter and accept the cost. Disputes here are settled with documentation, and a guarantee in writing is worth paying for.
If footfall feeds staff scheduling, opening hours, layout tests or campaign measurement, you are comparing periods rather than certifying totals. Consistent bias cancels out of a comparison, so a counter in the 90 percent range is genuinely adequate, and spending three times as much for two more percentage points changes no decision you will make with it. The one thing you cannot compromise on is consistency: a counter that is 90 percent accurate every day is far more useful than one that swings between 85 and 99 depending on the weather.
If the count feeds a safety or licensing occupancy limit, ignore per-crossing accuracy entirely and buy coverage instead. Every door counted, plus a reset procedure. A 99 percent counter on three of your four doors is worse than a 92 percent counter on all four, and that comparison is the one no datasheet will make for you. The same conversion discipline is worth applying to the online side of the business, where you can audit the conversion path on your website with the same before-and-after rigour you are about to apply to the store.
People Counting Accuracy Questions, Answered
What is a good people counting accuracy?
Retailers commonly hold a counter to 95 percent or better against a manual count before trusting conversion figures derived from it, and that is a reasonable bar for a main entrance. For trend and staffing decisions the 90 percent range is genuinely sufficient, because consistent bias cancels out when you compare one period against another.
Which people counting technology is the most accurate?
Overhead 3D stereo sensors are the accuracy leader, typically above 98 percent, because they see a controlled downward view of the threshold and can separate people standing close together. Break-beam counters sit at 85 to 95 percent, thermal around 85 percent, and software on existing security cameras at 70 to 90 percent depending mostly on mounting angle.
Why is my occupancy number wrong when my footfall number looks right?
Because they are calculated differently. Footfall is a cumulative total, so a steady counting bias just makes it consistently a few percent light. Occupancy is entries minus exits, so errors accumulate as a drift rather than averaging out, and a single uncounted door pushes it further off in one direction every day until a reset clears it.
Do people counters get less accurate when it is busy?
Yes, on every technology. Counting accuracy falls as people per minute rises, because groups walking abreast, occlusion and people pausing in the doorway all become more common. That is why an accuracy test run in a quiet hour flatters the system, and why any honest validation includes a peak hour.
How do I test my people counter against a manual count?
Stand at the door through one busy hour and one quiet hour with two clickers, counting entries and exits separately, then compare against the system for the same window. Record the error with a sign rather than as a bare percentage, log staff crossings separately, and finish with an empty-building occupancy reading at close to expose any uncounted door.
Can software on my existing cameras be an accurate visitor counting system?
It can, if the camera looks down at a clean threshold rather than across a room at head height. Geometry dominates: an overhead view of a controlled doorway out-counts a better algorithm looking across a crowded, sunlit atrium. The honest planning assumption is 70 to 90 percent on a typical security camera, then measure yours and find out where it actually lands.
Is a 99 percent accuracy claim from a vendor reliable?
Read the wording rather than the number. A guarantee backed by monitoring, as SenSource publishes at 98 percent, is a commitment. An up-to figure, as V-Count publishes at 99 percent, describes a ceiling under favorable conditions. Several well known vendors including Xovis and RetailNext publish no accuracy figure at all, which means a pilot in your building is the only evidence available.
Keep Reading
If you have settled the accuracy question and want to run counting on cameras you already own, see people counting software for existing CCTV cameras. To compare the specialist systems side by side, read the best people counting software roundup, and for the hardware budget see how much people counting sensors cost. Once the counts are trustworthy, people counting analytics covers turning them into staffing and layout decisions.
Measure It on Your Own Doorway
Connect an entrance camera, run the two-hour clicker test, and you will know your real accuracy figure by the end of the day. The free plan covers two cameras, which is enough to validate a door properly before you buy anything.