Dokument: Certified control for train sign classification
| Titel: | Certified control for train sign classification | |||||||
| URL für Lesezeichen: | https://docserv.uni-duesseldorf.de/servlets/DocumentServlet?id=69927 | |||||||
| URN (NBN): | urn:nbn:de:hbz:061-20250620-091457-0 | |||||||
| Kollektion: | Publikationen | |||||||
| Sprache: | Englisch | |||||||
| Dokumententyp: | Wissenschaftliche Texte » Artikel, Aufsatz | |||||||
| Medientyp: | Text | |||||||
| Autoren: | Roßbach, Jan [Autor] Leuschel, Michael [Autor] | |||||||
| Dateien: |
| |||||||
| Stichwörter: | Computer vision, Artificial intelligence, Formal methods, ATO, Autonomous systems | |||||||
| Beschreibung: | Certified control makes it possible to use artificial intelligence for safety-critical systems. It is a runtime monitoring architecture, which requires an AI to provide certificates for its decisions; these certificates can then be checked by a separate classical system. In this article, we evaluate the practicality of certified control for providing formal guarantees about an AI-based perception system. In this case study, we implemented a certificate checker that uses classical computer vision algorithms to verify railway signs detected by an AI object detection model. We have integrated this prototype with the popular object detection model YOLO. Performance metrics on generated data are promising for the use-case, but further research is needed to generalize certified control for other tasks. | |||||||
| Rechtliche Vermerke: | Originalveröffentlichung:
Roßbach, J., & Leuschel, M. (2025). Certified control for train sign classification. Science of Computer Programming, 246, Article 103323. https://doi.org/10.1016/j.scico.2025.103323 | |||||||
| Lizenz: |  Dieses Werk ist lizenziert unter einer Creative Commons Namensnennung 4.0 International Lizenz | |||||||
| Fachbereich / Einrichtung: | Mathematisch- Naturwissenschaftliche Fakultät | |||||||
| Dokument erstellt am: | 20.06.2025 | |||||||
| Dateien geändert am: | 20.06.2025 |
