Ausnutzung von Domänenwissen zur Absicherung Tiefer Neuronaler Netze
Abstract
This dissertation combines perception deep neural networks (DNNs) with domain knowledge, such as physical models, to identify safety-critical pedestrians. We introduce specialized metrics to assess pedestrian detections essential for safe driving. By incorporating planner-awareness into perception evaluation, we uncover safety-critical failure cases and enable mitigation strategies. These metrics provide a solid foundation for evaluating a model's safety compliance before deployment.
Fields of Study & Subjects
Autonomous Vehicle Technology and Safety
Advanced Neural Network Applications
Safety Systems Engineering in Autonomy
Publication details
- Journal / source
- mediaTUM – the media and publications repository of the Technical University Munich (Technical University Munich)
- Publication date
- 9 January 2026
- Language
- EN
- Document type
- Thesis