26 August 2026 · 2 min read

Counting store visitors with the cameras you already have

retail computer vision footfall

Almost every store has cameras. Almost nobody gets useful data out of them. Yet the most important question in physical retail — how many people walk in, and how many buy? — is answered exactly there, without buying anything new.

Why footfall changes decisions

Receipts tell you how many bought. Without traffic, though, you can't tell whether a sales dip is a footfall problem (marketing, window, hours) or a conversion problem (assortment, staff, pricing). Those are two different diseases with two different cures — and without counting entrances you're treating them blindfolded.

With visits and receipts cross-linked per store you get:

  • each store's real conversion, comparable over time;
  • the actual peak hours for planning shifts and openings;
  • the measurable effect of windows, promotions and re-layouts.

How it works without dedicated hardware

Commercial people-counters have existed for years: proprietary sensors, installation, subscription. The road I use is different: existing IP cameras already emit a video stream (RTSP) — a small server with a vision model analyses it and counts people in and out. No new sensors, no construction work.

Raw numbers lie happily, though: reflections, staff walking in and out, closing hours. That's why half the job is cleaning — filtering noise and anchoring the numbers to real opening hours. It's the difference between a counter and a number you can trust.

What about privacy?

The system counts silhouettes, it does not recognize people: no biometric data, no faces stored. It's a GDPR-compatible setup, documented in the store's privacy notice.

Where to start

With a pilot in a single store, on the cameras already there: two-three weeks to tune the model and verify accuracy against sample counts. Then extend across the network and cross-link sales — as in this real retail-network case.

Want to know if your cameras can do it? See the Video analytics & Computer Vision track or write to me — the camera model is all it takes to answer.

Read next