During the 4-week internship at Volvo Cars Ghent, I had my first exposure to the industry. Before the internship began, I underwent the necessary safety training. Following a guided tour of the welding shop, ideas and solutions were discussed. The project I was involved in focused on the glue placement check. Here a robot applied glue to car parts, which was then to be inspected using computer vision. Cameras were strategically placed to collect data. My primary task was to evaluate and manipulate this data and think out further steps in this project.

Supervisors

UNIVERSITY GHENT

dr. ir. Nele De Geeter

VOLVO CARS GHENT

Sam de Beul

Data preprocessing

A notebook was created to document the steps taken to preprocess the data and reduce noise. Since some image maps included multiple part positions, a function was developed to filter the images based on their positions in several maps. This approach can be applied to any new scenario without limitations on position views or objects. Various filters, including those involving masks, light, contrast, and multiple kernels, were tested. These tests were conducted for multiple cameras and positions.

Matching position views

Data augmentation

With transfer learning, state-of-the-art computer vision techniques were tested alongside simpler methods. Additionally, a pipeline and a data augmentation notebook were provided for further testing.