Uploaded August 2026 | Updated September 2026, 7 hours ago
This is the project video for the bachelor's thesis of Cedric Herren in the computer science department at the Bern University of Applied Sciences, Spring 2026.
Manually checking the completeness of playsets is a time-consuming and error-prone process, especially in toy libraries. Playmobil sets often consist of numerous individual components, and checking these upon return requires considerable manpower. Against this backdrop, this bachelor's thesis investigates how computer-based object recognition methods can be used to automate the completeness check of such playsets.
The aim of this work is to design, implement, and evaluate an application based on YOLO-based object recognition, integrating the entire workflow from data generation and model training to practical application. A key component of the solution is an automated labeling method that generates video-based training data using classical image processing techniques, significantly reducing manual annotation effort. The resulting datasets are used to train specialized YOLO models, each tailored to specific Playmobil sets.
This is the project video for the bachelor's thesis of Cedric Herren in the computer science department at the Bern University of Applied Sciences, Spring 2026.
Manually checking the completeness of playsets is a time-consuming and error-prone process, especially in toy libraries. Playmobil sets often consist of numerous individual components, and checking these upon return requires considerable manpower. Against this backdrop, this bachelor's thesis investigates how computer-based object recognition methods can be used to automate the completeness check of such playsets.
The aim of this work is to design, implement, and evaluate an application based on YOLO-based object recognition, integrating the entire workflow from data generation and model training to practical application. A key component of the solution is an automated labeling method that generates video-based training data using classical image processing techniques, significantly reducing manual annotation effort. The resulting datasets are used to train specialized YOLO models, each tailored to specific Playmobil sets.
