Why footfall counters produce numbers nobody trusts
Almost every retailer and venue operator has bought a people counter, and a surprising number quietly stopped using the number. It still appears in reports; nobody makes decisions on it. The reasons are specific and mostly avoidable, and knowing them is how you evaluate a counter properly rather than on a claimed accuracy figure.
Failure one: counting proximity instead of crossing
The most common flaw. A counting line is drawn across a doorway and the system counts anyone who comes near it. Someone standing by the entrance talking on the phone is counted repeatedly. A member of staff working the front table inflates the total all day.
The correct approach tests whether the person's path actually crossed the line: a true segment-intersection test between their movement vector and the line segment. Someone hovering beside a line never intersects it and is never counted, no matter how long they stand there.
How to test it: stand next to the counting line for two minutes without crossing, then check the count. This takes two minutes and eliminates a large share of products.
Failure two: counting people walking past
The mirror image. Someone walking parallel to the line, along a shopfront rather than into it, trips a proximity-based counter without ever entering. In a mall or a high street this is not an edge case, it is most of the traffic in front of your door.
A segment-intersection test rejects these too, because a path parallel to a line does not cross it, and a path that passes beyond the line's endpoint does not cross the segment even if it crosses the infinite line. That second distinction catches counters that get the maths half right.
Direction matters here as well. Taking direction from the sign of a cross product between the movement vector and the line stays correct at any approach angle, whereas simpler methods only work for people walking perpendicular to the doorway, which real customers rarely do.
| Scenario | Proximity counter | Intersection test |
|---|---|---|
| Walks in through the door | Counts | Counts |
| Stands beside the line talking | Counts repeatedly | Never counts |
| Walks parallel along the shopfront | Counts | Never counts |
| Passes just beyond the line's end | Often counts | Never counts |
| Enters at a shallow angle | Direction often wrong | Direction correct |
| Briefly hidden behind another shopper | Loses the track, misses the crossing | Holds the track through the gap |
Failure three: tracking the middle of the body
Most trackers follow the centre of a detected person. It is the obvious choice and it drifts badly in exactly the conditions a shop presents. When someone is half hidden behind a display, a pillar or another shopper, the detected box shrinks and its centre jumps sideways, which can register a crossing that never happened or lose one that did.
Anchoring the track at the feet is more robust, because the ground contact point is both what actually crosses a doorway and the part of a person whose position changes least when the rest of them is occluded. It is a small change with a disproportionate effect on accuracy in cluttered spaces.
This is also why accuracy figures measured in an empty test corridor do not transfer to a Saturday afternoon. Ask what the number was measured in.
Failure four: totals that live in memory
The quietest failure and the most damaging to trust. Many systems keep the running total in memory for speed. When the camera reconnects, the service restarts or the box reboots, the day's total is wrong, and nothing in the interface says so.
The fix is to treat the live total strictly as a cache and the event log as the source of truth. Every crossing is a stored event; totals are derived from those events and re-seed from them whenever a camera comes back. A restart then costs nothing.
There is a related design decision worth asking about. Violation-style events can safely be throttled to keep logs readable, but crossings never can. A dropped duplicate violation costs nothing; a dropped crossing silently corrupts the day's total, which is the entire product.
How should you actually test a counter?
With a clicker, over a fixed window, at your busiest hour. Have someone count manually for thirty minutes at the door during peak trade and compare. That single measurement is worth more than every accuracy claim in every brochure, because it is measured in your space with your traffic.
Then run the four deliberate tests: stand beside the line without crossing, walk parallel to it, walk past its end, and have two people cross together while one partially blocks the other. A counter that handles those four handles most of what a real doorway produces.
Finally, restart it mid-day and check whether the total survives. If it resets, the number was never trustworthy regardless of how accurate the detection was.
Common questions
- Why is my footfall counter counting too many people?
- Usually because it counts proximity to a line rather than testing whether the person's path crossed it. Anyone standing near the entrance is counted repeatedly, and people walking parallel along the shopfront are counted without entering. Test it by standing beside the counting line for two minutes without crossing and checking whether the count rises.
- How accurate should a people counter be?
- Ask instead what the figure was measured in. Accuracy in an empty corridor does not transfer to a busy shop floor, where occlusion is the dominant source of error. The only meaningful test is a manual clicker count over a fixed window at your busiest hour, in your own space.
- Why do footfall totals change or reset?
- Because many systems keep the running total in memory rather than deriving it from stored events. A camera reconnecting, a service restart or a reboot then loses or corrupts the day's figure with nothing in the interface indicating it. Totals should re-seed from an event log that is the source of truth, so a restart costs nothing.
- Does a people counter work when the shop is busy?
- It depends on how it tracks. Trackers following the centre of a detected body drift when someone is partly hidden behind a display or another shopper, which is precisely when accuracy matters most. Anchoring the track at the feet is more stable, because the ground contact point moves least under occlusion and is what actually crosses a doorway.
