People Counting Camera vs Overhead People Counter CCTV People Counting Software, Video Based People Counting, and How to Pick Between Them
Use an overhead people counter when the number becomes a denominator, and a people counting camera when the number is an operational signal. A dedicated sensor mounted straight down over the door reaches 98 percent or better because it sees the tops of heads and can split people walking abreast. CCTV people counting software runs on cameras you already own and lands around 70 to 90 percent, because a security camera is mounted high and at an angle, where bodies overlap. If a conversion rate, a lease negotiation or a fire-code occupancy figure depends on the count, buy the sensor. If you are staffing to traffic, watching queues or comparing sites, the software is accurate enough and costs a fraction.
People Counting Camera vs Overhead People Counter, Side by Side
Figures are US market ranges for 2026 at a normal single-door entrance. Every accuracy number in this category is a lab figure until it meets your own doorway.
| Factor | Overhead People Counter | People Counting Camera Software |
|---|---|---|
| Typical accuracy | 98%+ for 3D stereo | 70 to 90% |
| Hardware per door | $200 to $2,000 depending on technology | None, uses existing cameras |
| Installation | $100 to $500 per device, ceiling work | Connect a stream, draw a line |
| Recurring cost | $20 to $100+ per sensor per month, plus 8 to 12% support | Platform license only |
| Handles groups walking abreast | Yes, top-down view separates them | Weaker, bodies occlude at an angle |
| Time to cover 40 sites | Weeks to months of install scheduling | Days |
| Does anything besides counting | No, single purpose | Dwell, queues, search, security alerts |
| Good enough for conversion rate | Yes | Not without validation |
Why a Security Camera Counts Worse Than a Sensor Costing Less Than It
The gap is geometry, not intelligence. An overhead people counter is mounted straight down over the doorway. From directly above, two people walking shoulder to shoulder appear as two separate shapes with clear space between them, so splitting them is easy. That single mounting decision is most of the accuracy difference, and it is why a $900 sensor can out-count a $2,000 camera running a better model.
A security camera is mounted high on a wall and angled down the room, because it was installed to see faces and aisles, not to count crossings. From that angle a person walking behind another is partly hidden. The software has to infer that two overlapping shapes are two people, and sometimes it will not. Add a glass shopfront throwing a hard backlight for two hours every afternoon and the same camera that counts well at 10am counts badly at 4pm.
This is worth understanding before you judge a pilot. If video based people counting underperforms at your site, the fix is usually the view rather than the vendor. A camera repositioned to look down at the entrance, or one cheap camera added directly over the door and fed into the same software, often closes most of the gap for far less than a sensor program across every location.
The Test That Settles It: Will Anyone Divide by This Number?
Counting error behaves completely differently depending on what the number feeds. That, not the vendor shortlist, is the decision.
Buy the overhead sensor
The count is a denominator or a compliance figure. Conversion rate, sales per visitor, traffic evidence in a lease negotiation, or an occupancy limit a fire code depends on. Here a 10 percent undercount does not stay a counting problem: it silently inflates every derived metric, in every store, for as long as the system runs.
Use CCTV people counting software
The count is an operational signal and the shape of the curve carries the decision. Staffing to traffic, spotting queue build-up, comparing Saturday to Tuesday, ranking sites. Whether the true peak was 300 or 330 people, you make the same call, so paying four figures a door to remove that drift buys nothing.
A hybrid is common and usually sensible. Put dedicated counters on the two or three entrances that need audited numbers, and run software on every other camera for occupancy, dwell and flow. You pay for precision only where a decision depends on it. Where a chain-wide sensor rollout is being tracked as a multi-site capital program rather than an operating expense, that split also keeps the capital request small enough to actually get approved.
How to Validate Either One Before You Sign
Run the same test on whichever camp you pick, because published accuracy figures are measured on the vendor's doorway, not yours.
- 1Pick your busiest entrance, not your easiest one. A quiet side door will pass any test and tell you nothing.
- 2Hand-count a full busy hour, in and out separately. Directional error is the one that wrecks occupancy, and a net total can hide it.
- 3Compare against the same hour from the system. Retail buyers commonly hold a counter to 95 percent or better against a manual tally before trusting anything built on it.
- 4Repeat once in changed light. Late afternoon sun through a glass front is the condition that separates a real result from a demo.
- 5Walk the building and list every door a person can use. Any uncounted opening makes live occupancy drift in one direction all day.
Common Questions
Can a normal CCTV camera count people?
Yes, if it sees the doorway with a usable crossing line and outputs RTSP or ONVIF. Accuracy is typically 70 to 90 percent, lower than a dedicated overhead counter, because the camera views people at an angle where they overlap. Repositioning the camera to look down at the entrance usually improves it noticeably.
What is an overhead people counter?
It is a sensor mounted on the ceiling directly above a doorway, looking straight down. Stereo-vision, time-of-flight and thermal variants exist. The top-down view is what gives it accuracy, because people who walk in close together still appear as separate shapes from above.
How accurate is video based people counting?
Software on existing angled security cameras generally reaches 70 to 90 percent at a normal entrance. Purpose-built 3D stereo sensors reach 98 percent or better. Wide doorways, groups arriving together, and strong backlighting pull both numbers down, so validate on your own busiest door.
Is people counting software cheaper than sensors?
Substantially, because it removes the hardware and the install entirely. A sensor route carries $200 to $2,000 per door, $100 to $500 per device to install, $20 to $100+ per sensor per month, and 8 to 12 percent of hardware value per year in support. Software carries the platform license alone.
Does people counting need facial recognition?
No. Counting only requires deciding that a pattern of movement crossing a line is one person, and storing a number and a timestamp. No identity is involved. Face-based features are a separate product decision that brings its own state privacy questions, notably in Illinois and Texas.
Can I mix cameras and sensors in one system?
Yes, and for multi-site operators it is often the best answer. Put dedicated counters on the entrances that need audited numbers and run software on the remaining cameras for occupancy, dwell and flow. Just keep the two data sources labeled, because their accuracy is not comparable.
Related Reading
If you have landed on the software side, footfall counter software covers visitor counting and occupancy counting on cameras you already own, with the full cost table by technology. For the vendor shortlist across both camps, see the best people counting software roundup, and for hardware budgeting device by device, how much people counting sensors cost. The measurement question itself is covered in people counting accuracy, and retail teams wanting counts alongside loss prevention should read retail video analytics.
Test It on Your Own Doorway First
Connect an existing camera, draw a counting line and compare an hour against a hand count before you spend anything on sensors.