Project
befit
Two-part system that counts people through gym entrances. A GPU box inside each gym watches two cameras — one for entries, one for exits — recognises faces locally and keeps every biometric template on its own machine. What leaves the site is only a count: an anonymous id, a direction, a timestamp and optionally a gender and age bracket. The cloud service collects those counts from every site and makes them answerable in one dashboard. The privacy guarantee is structural rather than policy-based: the payload has no field able to carry a face vector and the cloud database has no column able to store one.
Local face recognition on a GPU box per site
Biometric templates never transmitted — enforced by schema, not policy
Entry/exit direction counting from two cameras
Optional gender and age-bracket reporting, degrading to 'unknown'
Batch reporting with heartbeats and retry tolerance
Central dashboard aggregating every site