How do you train your AI?
Aura Vision fine-tunes its AI to each store using a small set of face-blurred snapshot images during the first week of operation, then continuously improves through quarterly accuracy audits. No personal data is ever used.
Aura Vision fine-tunes its AI to each store using a small set of face-blurred snapshot images during the first week of operation, then continuously improves through quarterly accuracy audits. No personal data is ever used.
What this means in practice
- No facial recognition. Faces are blurred at source on the APU before any image leaves the device. The blurring is irreversible — the original face cannot be recovered.
- No personal data. A blurred snapshot doesn’t identify the person. Locations and timestamps of snapshots are not stored.
- Snapshots are deleted. Training data is held only briefly during the tuning window, then discarded.
- Tuning is per-store. The AI is adapted to your store’s lighting, camera angles, staff uniforms and customer demographics — so accuracy is high from day one rather than relying on a generic model.
How it works
The Computer Vision team uses the blurred snapshots to refine three things for your store: people counting accuracy, age and gender estimation accuracy, and staff-vs-customer recognition (matching your team’s uniform). Once tuned, the model runs entirely on the on-premise APU — only anonymous, aggregated analytics ever reach the cloud.
The same audit-and-tune loop runs quarterly on a randomised sample of stores to maintain accuracy as cameras, layouts and uniforms change.
Where to read more
- Product privacy — privacy by design
- Security at Aura Vision — full architecture
- Does it use facial recognition?
- Is the data secure and compliant with GDPR?