The most-copied sentence in counting RFPs
Counting tenders routinely carry a line like "the system must achieve at least 95% counting accuracy". It sounds rigorous and specifies almost nothing, because the sentence omits everything that determines whether the number is meetable or even measurable: accuracy of what, measured how, over what period, under what traffic conditions, compared against what reference. Vendors know this, which is why nearly every bid claims compliance and why acceptance disputes are common.
What does "95% accuracy" actually mean in people counting?
It depends on the definition the specification chose, and most specifications choose none. It might mean net accuracy (total counted versus total actual over a period, where overcounts and undercounts cancel and flatter the result), absolute or gross accuracy (errors summed regardless of direction, a much harder and more meaningful test), directional accuracy (in and out counted correctly, which matters for occupancy), or per-interval accuracy (each hour within tolerance, the strictest and most operationally relevant). A vendor can pass net accuracy and fail per-interval accuracy at the same site with the same data. Specify which one you mean, or you have specified nothing.
Write the number with its conditions attached
A testable accuracy requirement names five things: the metric definition (from the list above), the reference (a manual count by a named method, or a video-verified ground truth), the measurement window (a continuous period covering real peaks, not a quiet Tuesday), the traffic conditions (including the busiest hour and group arrivals, since crowding is where cheap sensors fail), and the tolerance for known-hard cases (staff movements, groups entering abreast, prams and trolleys). The counting-specific error sources are catalogued in people counting accuracy factors, and the group problem in group entry counting.
The specification lines buyers most often forget
- Staff exclusion. Uncounted staff movements inflate visitor numbers and corrupt every conversion metric downstream. Specify how staff are handled and how it will be verified.
- Peak-hour performance. Accuracy at 20 people per minute is a different engineering problem than accuracy at 2. Specify the peak that matters.
- Sensor placement constraints. Mounting height ranges and coverage per unit determine how many devices your doors need; a bid that under-counts devices will meet the price criterion and fail the accuracy one. See ToF accuracy and range in practice.
- What "counting" includes. Entries only, or entries plus dwell plus zone paths? Scope creep between specification and expectation is the most common source of post-award disappointment.
Do not accept anyone's datasheet, ours included
The honest position for a buyer is that no vendor's advertised accuracy figure, Ariadne's included, transfers to your doorway without verification, because the number was measured under conditions that are not yours. Write the acceptance test into the contract, run it on your own doors, and make payment or renewal contingent on it: the procedure is in the accuracy test methodology and the contractual framing in acceptance testing. A vendor who resists a written on-site test has told you something useful.
The full question set for the rest of the specification lives in the RFP template.
---