Digital solutions
HIGH-PRECISION, LOW-LATENCY INDUSTRIAL COMPUTER VISION AT THE EDGE
The difference between an AI demonstration and a production-grade system is measured on the ground. A license plate recognition model trained on generic datasets collapses when facing local Tunisian plates, misaligned cameras, heavy rain, or glare at sunset.
Our approach is empirical. We start from your real video streams, build an annotated dataset directly from your operational environment, and train a dedicated model.
Field Deployment: Bizerte Economic Activity Park
On the Bizerte economic activity park project, this methodology increased recognition accuracy from 50 - 65% with a generic model to approximately 95% using a model specifically trained on local Tunisian license plates, based on 3,200 annotated images.
Mastery of Operational Edge Engineering
We address the most critical and often overlooked technical challenges: handling multiple simultaneous RTSP streams, optimizing latency, configuring local GPU hardware, system supervision, and ensuring graceful failover when a camera disconnects.
Multi-Stream Ingestion & Low Latency: High-throughput decoding of concurrent HD video streams with zero frame loss via native GPU/Edge optimization.
System Resiliency & Fault Tolerance: Real-time camera health monitoring and instant failover mechanisms in case of hardware or network outages.
Data Sovereignty & Security: 100% On-Premise / Air-Gapped architectures ensuring complete data privacy and enterprise compliance.
Automatic number plate recognition (ANPR / LPR)
Model training on annotated local data
Object detection and tracking (YOLO)
OCR and document data extraction
Real-time processing of multiple RTSP streams
Sizing and installation of dedicated GPU workstations
Integration of models into an existing product
Systems already delivered in this scope.
Economic activity park
PAEB — Parc d'Activités Économiques de Bizerte
A plate recognition system driving three barriers from six video streams, with a model trained specifically on Tunisian plates.
Logistics
It depends on how variable the conditions are. For the Tunisian plate model on the Bizerte project, 3,200 annotated images were enough to reach approximately 95% accuracy. A site with more camera angles or harder lighting will need more.
For real-time video, almost always on-premise. Pushing several RTSP streams to the cloud is expensive in bandwidth and adds latency that is incompatible with opening a barrier. We size a dedicated GPU workstation on site and send only the results upstream.
Tell us the scope. We reply as soon as possible.
