Artificial intelligence to improve number plate recognition and ticket processing’une meilleure lecture des plaques d’immatriculation et du traitement des contraventions

Submitted on 15/11/2024
image décorative
image décorative

Artificial intelligence to improve number plate recognition and ticket processing

Florent Meyer, a graduate of the Université de Technologie de Compiègne (UTC), joined IRISA in 2023. He is a doctoral student on a CIFRE thesis within the SHADoc research team and ANTAI, the public body in charge of the automated processing of offences.

The subject of his thesis is image and text processing, and more specifically the construction of an externalised language model in a machine learning system for number plate recognition.

With more than 16 million offences recorded by radar, the performance of automatic number plate recognition is crucial

In 2022, more than 16 million offences were recorded by radar: with such a volume, the performance of automatic number plate reading is crucial for ANTAI, but it is limited by two factors.

  1. On the one hand, reading performance decreases with the constant evolution of number plates in vehicle fleets, both in France and abroad. This is because the order, distribution and number of letters and numbers on the plates vary over time. On French plates, for example, the letter H will appear at the top of the registration plate for the first time at the end of 2024. Adapting to these changes now requires frequent and costly retraining of the reading systems.
     
  2. In addition, external factors such as weather conditions, vegetation in front of the cameras or optical distortion can hamper reading. Current systems struggle to identify illegible characters on images that are too degraded, and risk offering incorrect readings. This leads to an increased need for plate reading by operators.

Florent's research work consists of ...

Florent's research aims to overcome the limitations of current approaches. He is attempting to design a machine learning model capable of recognising characters and indicating those that cannot be read, while making it easier to update as the vehicle fleet changes. The concepts of adaptability, rejection capacity and interpretability are at the heart of his research, with the aim of creating robust, high-performance and self-evolving systems.

Florent's thesis will help ANTAI to improve the processing of fines by reducing the propagation of errors in the analysis of number plates, while saving time, reliability and speed. This will also help to guarantee equal treatment for all vehicles travelling in France.

And in conclusion: drive carefully!