Vision AI Face Recognition
Contactless, AI-driven face recognition with on-premise edge inference.
We enable AI for trusted identification
Our Vision AI Face Recognition verifies members contactlessly in under a second. Full data sovereignty through on-premise processing.
This page focuses on the technical aspects of the AI solution: edge inference on NVIDIA Jetson, a multi-stage AI pipeline with authenticity / liveness checks, fully local data handling without cloud. For fitness-specific benefits, the commercial model and pricing, please visit our product page gymedgeofficial.de.
Technical cornerstones
Edge AI on Jetson, dual-camera with authenticity check, fully on-premise processing.
Contactless recognition in under one second – even under varying lighting
Multi-stage anti-spoofing pipeline with dual-camera (color + infrared)
Real-time inference on NVIDIA Jetson Orin Nano – edge instead of cloud
Fully on-premise data – images, vectors and logs never leave the studio
Pre-trained models – no training on customer data
Mathematically non-reconstructable face vectors instead of raw images
More than 50 checks per second with a tiered filter pipeline
Right to erasure at the click of a button – GDPR Art. 17 compliant
The AI pipeline in detail
Five stages, fully local on the edge device – every stage can early-exit to minimize latency and energy use.
Face detection – locating the face in the live camera feed
Alignment & head pose – normalization, abort on extreme pose
Authenticity check – rejects photos, screens and masks
Vector computation – mathematical fingerprint (embedding)
Matching – comparison against the local vector database
Performance at a glance.
End-to-end latency from camera to access decision
Live stream is checked many times per second
Data packets sent to cloud or third-parties in normal operation
Edge compute instead of cloud
Full inference on an edge device – directly at the studio, with no internet dependency.
The entire processing chain – detection, authenticity check, vector computation and matching – runs on a compact NVIDIA Jetson Orin Nano edge device at the studio. Power draw between 7 and 15 watts, latencies under one second, no cloud round-trips.
The check-in terminals are stateless: they stream camera frames to inference and display the result. No personal data remains on the endpoint device. Database, vectors and image archive reside exclusively on the edge server's local SSD.
The AI models are pre-trained on public research datasets. There is no training and no fine-tuning on member data – this avoids any coupling between model weights and individuals, and new installations are ready to operate immediately.
Data sovereignty & GDPR
Data processing at the studio – not in the cloud.
Biometric data does not leave the studio. There is no transmission to external AI providers, cloud services or third parties. The stored vector (mathematical embedding) does not allow reconstruction of a face image – vector computation is a one-way operation.
Members actively confirm the privacy policy, terms and right of withdrawal before registration. Retention periods, automatic clean-up runs and the right to erasure (GDPR Art. 17) are implemented natively. Legal and contractual framework set up in cooperation with the law firm Dr. Geisler, Dr. Franke Rechtsanwälte Partnerschaft mbB.
The commercial and contractual track for studios is handled under the sales brand GymID. All packages, terms and a personal demo appointment are available via gymedgeofficial.de.
Frequently asked questions
Technical answers about the Vision AI Face Recognition.
The entire AI pipeline (detection, authenticity check, vector computation, matching) runs on a local NVIDIA Jetson Orin Nano edge device at the studio. No image or biometric data is ever transmitted to cloud services or external AI providers.
End-to-end latency from camera frame to access decision stays below one second. The live stream is processed at over 50 checks per second. A tiered filter pipeline early-exits each stage when downstream steps would be unnecessary – saving latency and energy.
The terminals use a dual-camera combination of color and infrared sensors. Photos and screens look fundamentally different from real faces in the infrared spectrum, making reproductions immediately recognizable. The model also detects mask materials. Sensitivity thresholds are configurable per camera type.
On the edge server at the studio: master data (synchronized from the CRM if applicable), face images, mathematical face vectors and audit logs. The check-in terminals themselves are stateless – they stream camera frames for processing and display results, but store no personal data locally.
No. The vector (embedding) is the result of a one-way operation by the neural network. The vector cannot be inverted into a face image. Even on a database leak, the vectors are useless without the matching model and without inversion attacks.
No. Models are pre-trained on public research datasets and operated as shipped. There is no training and no fine-tuning on member data. This avoids any coupling between model weights and specific individuals.
For each member, all personal data can be deleted at the click of a button: master data, face images, all vectors, all audit entries and archived raw images. Scheduled deletions for a future date are also possible. Retention periods for audit logs and raw images are configurable.
Sales, terms and implementation are handled by our partner under the GymID brand. A complete overview including a demo request is available at gymedgeofficial.de.
Let's enable AI together!
Talk to our AI experts about the technical integration of Vision AI Face Recognition. For fitness-specific inquiries please visit gymedgeofficial.de.
When you click "Schedule via HubSpot", data will be transmitted to HubSpot.
By submitting your data, you agree to the processing. For more information, see our Privacy Policy.
