In road traffic, safety is a “must-have”. This is why motorized vehicles are regularly put to the test in almost all European countries. Do the engine and brakes function properly? Do the wheels have sufficient profile? Is the chassis OK? All this is carefully checked. In addition, test centres and workshops also check the authenticity of vehicle documents and registration plates.

AI prototype for process automation in the automotive environment

The vehicle identification number (VIN) plays an important role in this context: it is as unique as a fingerprint and provides important information on vehicle history and equipment. At present, VINs are still generally recorded manually in workshops, test facilities, and road traffic offices. This bears the risk of errors and also costs time. This process can be fully automated with the help of artificial intelligence and machine learning (ML). We have tested how this can work on behalf of a European platform provider. The result is a prototype for the automatic recording of chassis and master numbers – and all within a development period of only three weeks.

Our AI solution paves the way for road traffic authorities, garages, and test centres to achieve significant time savings and quality improvements – and is also very easy to use. All the user has to do is take a picture of the vehicle registration document on a tablet, and the artificial intelligence does the rest: it extracts the 17-digit VIN from the photo and then transfers it to the desired IT system in the back end. The advantage of this solution: transposed numbers and transmission errors are a thing of the past, the captured VIN is available at the touch of a button.

Sounds pretty simple, actually. But the road to getting there is an enormous challenge – at least for artificial intelligence: It must learn to find the right information from the multitude of data in the vehicle paper and then transform the image into a column of numbers. And that is easier said than done. Because normal neural networks (NN) can detect given objects within an image. However, if the position of the object changes, many NN fail to locate it. In addition, different image qualities resulting from different shooting angles, lighting conditions and distances make it even more difficult to identify the desired information. For while the human brain automatically suppresses this “noise”, intelligent software must learn from scratch to correctly classify such entropies.

To make learning easier for our AI prototype, we have developed the model on the basis of two Convolutional Neural Networks (CNN). The Deep Learning Architecture is specialized in image processing and uses different filters (Convolutional Layer) and aggregation layers (Pooling Layer). These enable CNN, in contrast to conventional NN, to recognize displayed images based on a given matrix and to transform them into the desired data format.

Software model masters complex tasks with confidence

So in our case, the task was: Find the VIN in the photographed vehicle paper and transform it into a 17-digit number. We had to train artificial intelligence accordingly. Ideally, this would require an extensive database. The larger and more varied the training pool, the higher the hit rate in the end. However, for data protection reasons, we only had 40 (mainly expired) registration papers – far too few to show the required amount of deviations and entropies in training. So we artificially enlarged our database so that it ultimately contained around 3,000 different virtual vehicle papers.

Now nothing stood in the way of iterative training. In concrete terms, this means that we continuously fed the artificial intelligence with new learning data, evaluated results and, on this basis, modified and improved the CNN matrix again and again. A complex process that took a lot of time: In the current ERNI prototype, the iterative training of artificial intelligence accounted for around two-thirds of the total development time. An effort that was worthwhile, however: In the end, the software delivered the correct VIN in nine out of ten cases – and that after only two weeks of training.


Wicher Visser
Principal Consultant
ERNI Switzerland

Marek Linka
Senior Software Developer
ERNI Singapore

News from ERNI

In our newsroom, you find all our articles, blogs and series entries in one place.

  • 27.09.2023.

    Unveiling the power of data: Part III – Navigating challenges and harnessing insights in data-driven projects

    Transforming an idea into a successful machine learning (ML)-based product involves navigating various challenges. In this final part of our series, we delve into two crucial aspects: ensuring 24/7 operation of the product and prioritising user experience (UX).

  • 13.09.2023.

    Exploring Language Models: An overview of LLMs and their practical implementation

    Generative AI models have recently amazed with unprecedented outputs, such as hyper-realistic images, diverse music, coherent texts, and synthetic videos, sparking excitement. Despite this progress, addressing ethical and societal concerns is crucial for responsible and beneficial utilization, guarding against issues like misinformation and manipulation in this AI-powered creative era.

  • 01.09.2023.

    Peter Zuber becomes the new Managing Director of ERNI Switzerland

    ERNI is setting an agenda for growth and innovation with the appointment of Peter Zuber as Managing Director of the Swiss business unit. With his previous experience and expertise, he will further expand the positioning of ERNI Switzerland, as a leading consulting firm for software development and digital innovation.

  • data230.08.2023.

    Unveiling the power of data: Part II – Navigating challenges and harnessing insights in data-driven projects

    The second article from the series on data-driven projects, explores common challenges that arise during their execution. To illustrate these concepts, we will focus on one of ERNI’s latest project called GeoML. This second article focuses on the second part of the GeoML project: Idea2Proof.

  • 16.08.2023.

    Unveiling the power of data: Part I – Navigating challenges and harnessing insights in data-driven projects

    In this series of articles (three in total), we look at data-driven projects and explore seven common challenges that arise during their execution. To illustrate these concepts, we will focus on one of ERNI’s latest project – GeoML, dealing with the development of a machine learning algorithm capable of assessing road accident risks more accurately than an individual relying solely on their years of personal experience as a road user, despite limited resources and data availability.


  • 09.08.2023.

    Collaborative robots revolutionising the future of work

    The future of work involves collaboration between robots and humans. After many years of integrating technology into work dynamics, the arrival of collaborative robots, or cobots, is a reality, boosting not only safety in the workplace but also productivity and efficiency in companies.

  • 19.07.2023.

    When the lid doesn’t fit the container: User Experience Design as risk minimisation

    Struggling with a difficult software application is like forcing a lid onto a poorly fitting container. This article explores the significance of user experience (UX) in software development. Discover how prioritising UX improves efficiency and customer satisfaction and reduces risks and costs. Join us as we uncover the key to successful software applications through user-centric design.

  • 21.06.2023.

    How does application security impact your business?

    With the rise of cyber threats and the growing dependence on technology, businesses must recognize the significance of application security as a fundamental pillar for protecting sensitive information and preserving operational resilience.