In the world of retail analytics, accuracy starts with one key principle: filtering out staff from customer footfall. At FootfallCam, we’ve taken a privacy-first, AI-driven approach to solving this challenge — using Re-Identification (Re-ID) technology to exclude staff without the need for badges, wearables, or manual interaction.
This post explains the rationale behind this method, how we train our models, and the proven accuracy behind our solution — already deployed in thousands of stores worldwide.
Why Staff Exclusion Matters
In any store, a significant portion of daily footfall is from staff — not customers. Store associates walking in and out, merchandisers doing planogram checks, cleaners, and security personnel all show up in traditional people counting systems. Without proper exclusion, this skews your key retail KPIs:
- Inflated conversion rates
- Misleading zone-level dwell time
- Incorrect peak hour analysis
- Ineffective heatmaps and journey paths
The result? Bad data leads to bad decisions.
That’s why FootfallCam uses Re-ID — a computer vision model that recognises repeating movement patterns over time to distinguish staff from customers, passively and anonymously.
How Re-ID Works in Retail
Re-ID doesn’t rely on facial recognition or personal identifiers. Instead, our system uses non-PII visual features — body silhouette, outfit consistency, walking gait — to build a unique movement signature.
When someone reappears across different zones or re-enters the store multiple times in a day, the system flags them as a likely staff member. Over time, these signatures become more reliable, enabling automated exclusion from analytics dashboards and exports.
And because this is fully edge-processed, no sensitive footage leaves the store, and all identities remain anonymous.
Model Training and Accuracy
At FootfallCam, we’ve trained our Re-ID engine using over 2 million retail video segments across supermarkets, fashion chains, convenience stores, and QSRs — accounting for variations in lighting, layout, ceiling height, and uniform policies.
It took 3-6 months of iterative training and in-field validation to tune the model across different store formats. Today, we benchmark accuracy at:
- 94–97% precision in correctly excluding known staff
- Less than 2% false exclusion of actual shoppers (often staff shopping off-shift)
The system improves over time through continuous learning, adjusting to new uniforms, seasonal outfit changes, and store-specific routines.
Re-ID is the only method that scales effortlessly across a retail chain, requires no behavioural change, and works silently in the background.
Privacy, Always
Our Re-ID system is fully GDPR-compliant. No images are stored. No person is identifiable. Just pattern-matching of an anonymous movement — a clean, ethical solution to a long-standing problem.
#footfallcam #peoplecounter #peoplecounting #staffplanning #staffmanagement #retail #smartretail #aianalytics #reidentification #reiidtechnology
