As part of its commitment to innovation, Numeryx has developed TollXpress, a smart toll system based on artificial intelligence. We met Ahlem Barhoumi, Senior AI Engineer at Numeryx, to discover the behind-the-scenes of this ambitious project — from the initial idea to the first concrete successes.
Q1. Ahlem, how did the TollXpress project idea come about and what vision did Numeryx have behind this smart toll system?
It all started from a simple but ambitious idea: making toll collection smoother, faster and fully automated thanks to AI. At Numeryx, we have always wanted to transform traditional infrastructure into smart solutions, able to adapt to local needs and optimise the user experience.
The objective of TollXpress was therefore clear: to automate the classification and pricing of vehicles according to their type, number of axles and the nature of their cargo. Beyond the technology, it is a visionary project, born of a spirit of innovation and a desire to have a concrete impact on mobility and road management, in Africa and beyond.
Q2. You mentioned that no data existed at the start. How did you approach this challenge and lay the first foundations of the project?
Indeed, it was one of the biggest challenges of the project: no database was available at the start. So we had to be very resourceful. I began by researching and collecting specialised public data, then we gradually enriched our database with real images of vehicles detected in the field.
To strengthen the robustness of our models, we used data augmentation techniques: rotation, angle changes, added noise… to imitate real conditions (rain, shadow, dust). It was a long and demanding phase, but it revealed the creativity and rigour of the team. It is really this groundwork that allowed TollXpress to lay solid and reliable foundations.
Q3. Why did you choose YOLOv8 and transfer learning for this project, and what made their implementation so decisive?
We chose YOLOv8 for its speed and accuracy: it can detect several objects in an image in real time — essential for a toll system.
We then applied transfer learning, a technique that consists of adapting a model already trained on millions of images such as (COCO) to a specific task: here, axle detection, cargo classification and tricycle recognition.
Thanks to this approach, we achieved strong results from the start, while demonstrating our expertise in applied AI. This choice perfectly illustrates Numeryx’s philosophy: combining innovation and pragmatism to create high-impact solutions.
Q4. The project faced significant structural and hardware constraints, notably related to GPUs. How did you manage to optimise and deliver a prototype in just 4 days?
Yes, it was a real race against the clock! We did not have access to NVIDIA GPUs, which are often essential for this type of project. Rather than holding us back, we took it as an opportunity for innovation.
We converted our models to make them compatible with Intel GPUs, optimising every parameter to maximise performance despite the constraints. Thanks to this agile approach, we were able to deliver a first working prototype in just four days, from 14 to 18 April 2025.
It is a fine example of responsiveness, innovation and adaptability — three values at the heart of the DNA of Numeryx and TollXpress.
Q5. Today, what concrete results have you achieved with TollXpress, and how do you see the project evolving in the coming months?
The results are already very encouraging. TollXpress now reaches about 70% accuracy, while maintaining real-time processing, which is essential for a toll system.
We are now continuing with a dynamic of continuous improvement, notably through the integration of Vision-Language Models (VLM) and OpenAI’s CLIP model to strengthen tricycle detection — a real challenge in some countries.
Our goal is clear: to make TollXpress a complete, modular AI solution adapted to local specificities, able to evolve at the pace of market needs. This project goes beyond the purely technical dimension: it represents a key step towards smart transport, where AI becomes a true driver of efficiency, sustainability and innovation.
Each step of this adventure has been a learning experience and a stimulating challenge: lack of data, technical constraints, diversity of vehicles… so many opportunities to reinvent our methods and push the limits of AI applied to transport.
With TollXpress, we are not just creating an automated toll system, but a solution able to transform mobility, improve safety and efficiency on the roads, while paving the way for new innovations.
