Skip to main content
  • Home
  • References
  • Partners
  • About
Contact us
Provesta Soft

We design, deliver and operate digital platforms, AI systems and infrastructure for institutions and companies.

[email protected]WhatsApp : +216 94 443 455

Expertise

  • Services
  • References
  • Partners
  • Gallery

Company

  • About
  • Process
  • Blog
  • Careers
  • FAQ
  • Contact us

Where we work

  • Tunisia
  • Congo - Kinshasa
  • France
  • United Kingdom
  • United Arab Emirates
  • Egypt
  • Saudi Arabia
  • Algeria
  • Kenya
  • Benin

© 2026 Provesta Soft. All rights reserved.

  • Privacy policy
  • Legal notice
  • Cookie settings
  1. Home
  2. References
  3. Access control by automatic plate recognition

Economic activity park

PAEB Bizerte: automated plate-based access control

A plate recognition system driving three barriers from six video streams, with a model trained specifically on Tunisian plates.

Organisation
PAEB — Parc d'Activités Économiques de Bizerte
Sector
Logistics
Country
  • Tunisia
Contextual photograph: an outdoor surveillance camera. Awaiting a real capture of the PAEB system.

Measured figures

~95%
Recognition accuracy

Model trained on Tunisian plates

50-65%
Generic model benchmark

Same cameras, same conditions

10000+
Annotated training images

Collected on site

6
Video streams processed in parallel

RTSP, dedicated GPU workstation

3
Barriers controlled

The problem

Challenge. Secure vehicle access to an economic activity park without physical badges, with reliable passage traceability.

The solution

Our work. Deployment of an ANPR system: camera streams, plate reading, barrier control and real-time access history for PAEB operations.

  • Automatic plate recognition
  • Passage supervision and logging
  • Barrier integration and access scenarios

The result

The model trained on Tunisian plates reaches approximately 95% recognition accuracy, against 50-65% for a generic model under the same conditions. The three barriers are driven automatically from the reads, and every movement is recorded with its timestamp and the vehicle image.

Why a dedicated model

A detection model learns the appearance of what it has seen. Tunisian plates differ from European ones in format, typography and the presence of Arabic characters; a model that has never seen them fails predictably. Building a local dataset is an upfront cost, but it is the only lever that genuinely moves accuracy.

On-site processing rather than cloud

Six continuous video streams are a volume you do not push over an internet link without consequences for cost and latency. Processing locally means a barrier opens without waiting for a network round trip, and the system keeps working if the connection drops.

Technical stack

  • YOLO
  • OCR
  • Python
  • PyTorch
  • OpenCV
  • ANPR / LPR
  • RTSP
  • Laravel
  • WebSocket
  • Linux

Services involved

  • Custom web and mobile platforms
  • Applied computer vision and AI
  • Cloud, DevOps and infrastructure

Related organisations

  • Bizerte Economic Activity Park

More of our work

  • Contextual photograph: a computer classroom. Awaiting a real capture of the PAAF / FOADE Moodle platform.

    Ministry of Education programme, DRC

    National distance learning platform for teachers

    Ministère de l'Éducation — PAAF / FOADE

    Complete customisation of a Moodle platform for a national teacher training programme in the Democratic Republic of the Congo, with production, test and staging environments.

    Education

    • Congo - Kinshasa

    Read more

  • Contextual photograph: a taxi in the city. Awaiting a real capture of Winween.

    Ride-hailing operator

    Ride-hailing, loyalty and a deals marketplace

    Winween

    The super-app that brings value back to local communities. Ride-hailing drivers, shopkeepers, artisans, and service providers: Winween brings local economic players together in a single app. 0% commission, €0 subscription fee, and direct payment between customer and partner. France, not Silicon Valley.

    Ride-hailing

    • France

    Read more

Let's talk about your project

Describe your need: an engineer replies within 48 business hours.

Describe my needReview the references