Too bright or too dark... With car headlights, both options are dangerous. But not all headlights are the same.
The settings are becoming increasingly sophisticated. Testing devices are used for the approval and regular inspection of headlights in the automotive sector. These use sensors to record the intensity distribution of the headlamp light and evaluate it according to specified evaluation procedures.
New technologies such as matrix LED headlamps and a large number of new manufacturer-dependent “Advanced Frontlighting System” (AFS) light distributions pose major challenges for existing evaluation procedures. They also require workshop mechanics to have ever more extensive knowledge of how to use the test equipment.” This is what it says in the technical article of the Mittelstand Digital Zentrum Chemnitz about the project “Classification of headlights using AI”.
We are part of the project and your contact for software development and image processing.
The project and initial results were presented at the SpectroNet Collaboration Conference 2023 in Erfurt. These already show the high potential of AI in the field of image processing and sensor technology and are an example of AI projects that we can implement for our customers.
The increasing complexity of headlight types makes it difficult to classify them and cannot be mastered using analytical methods alone. AI methods are now to supplement or replace these. Choosing the right algorithms, hyperparameters and training methods is crucial. AI models must be able to recognise and interpret subtle differences and complex patterns in the data. Furthermore, effectively processing the images and extracting the features relevant for classification is technically challenging.
Firstly, the exact objectives, the requirements for the AI solution and the current data situation were specified in the onboarding workshop. The company partner presented its use case in detail. The team then carried out a comprehensive data analysis and pre-processing. Next, a simple AI model focussing on the classification task was created, tested, continuously expanded and adapted to the required complexity.
The team from the Mittelstand-Digital Zentrum Chemnitz developed a prototype of a complex AI model that lies within the predefined accuracy range. The project partners placed particular emphasis on the explainability and transparency of the industrial AI solution. The findings and optimisations to date have been comprehensively documented and integrated into a demonstrator application for the headlight testing software.
"The collaboration with the Mittelstand-Digital Zentrum Chemnitz was a real benefit for our company. Together, we have developed an innovative AI solution for classifying headlights that not only increases our efficiency, but also sets new standards in quality assurance. This partnership has shown us how valuable the use of AI in production is."
Dr.-Ing. Uwe Schleichert, Managing Director of VELOMAT Group
Deutsch 
español 
Ελληνικά