Retail Analytics Terminal
Embedded AI-powered footfall and dwell-time terminal with a live analytics SaaS dashboard.
What we delivered (snapshot)
- Embedded terminal (Raspberry Pi CM4 + custom carrier PCB) with on-device ML inference for footfall and dwell.
- Edge AI pipeline using MobileNetV2-based people detection, running at 10 FPS on-device.
- Multi-tenant SaaS dashboard with per-zone analytics, heatmaps, and CSV/API export.
The Challenge
A retail property management SME needed anonymous footfall and dwell-time analytics across 6 retail units without storing any identifiable footage — a hard requirement for GDPR compliance. Existing CCTV analytics vendors were ruled out due to data residency concerns.
The solution had to run all AI inference on-device (no video leaving the premises), aggregate only anonymised counts and timestamps to the cloud, and present this to property managers in a simple multi-site dashboard accessible on any device.
Constraints
Timeline
16 weeks to a 2-site pilot
Budget Band
£35k–£55k (hardware + AI + dashboard)
Technical Constraints
- All video inference must run on-device — no video uploaded
- GDPR-compliant: only anonymised counts leave the device
- On-device inference at ≥ 8 FPS on CM4
- Wi-Fi or 4G fallback for connectivity
Our Approach
- 1
Fit check & constraints clarification
Reviewed GDPR requirements, confirmed on-device inference approach, selected CM4 with Google Coral TPU for AI acceleration.
- 2
Requirements & acceptance criteria for V1
Defined counting accuracy target (≥ 90%), inference FPS floor (8 FPS), and data schema for the dashboard.
- 3
Architecture & risk register
On-device pipeline architecture, cloud aggregation design, risk flagged on camera FoV calibration per unit layout.
- 4
Design & build
Carrier PCB for CM4, on-device AI pipeline in Python/TFLite, React SaaS dashboard, and device management API.
- 5
Integration & validation
2-site pilot installation, calibration procedure, accuracy validation using manual ground-truth counts.
- 6
Handover package
Hardware files, AI pipeline repo, dashboard source, device provisioning guide, and recorded handover.
What We Delivered
- CM4 carrier PCB schematic and layout
- Gerbers, BOM, and assembly notes
- Power management and PoE integration notes
- Linux device firmware image (CM4)
- On-device AI pipeline source (Python + TFLite)
- Device management API and OTA update module
- Wall-mount enclosure CAD
- Camera bracket and aiming assembly
- FDM-printable prototype files
- React multi-tenant SaaS dashboard source
- Node.js aggregation API
- Device provisioning guide and admin documentation
- Heatmap generation module documentation
Outcome
"Pilot deployed across 2 retail units. Footfall counting accuracy validated at 91% against manual ground-truth counts. The SME presented the analytics to their retail tenants within 2 weeks of go-live and approved a 6-site roll-out."
Handover
- Hardware, firmware, and dashboard repos
- Device provisioning and calibration guide
- SaaS platform deployment runbook
- AI model files and retraining instructions
- Recorded handover session
Related Services
Start Your Project
Ready to build your prototype?
Tell us about your idea and we'll help you plan the fastest path to a working prototype or V1.