Towards the certification of vision based systems: modular architecture for airport line detection
Résumé
This paper addresses the certification issues for vision based systems embedded on civil aircraft to assist pilots during the ground navigation phase. We propose a design methodology, through the example of a modular architecture to detect ground mark lines in a color image. Our detection method is based on the combination of classical image processing algorithms, deep learning methods and a particle filter algorithm. We argue that the main interest of this architecture is to ease the certification problem when compared to an end-to-end neural network. We discuss about difficulties around certification and propose some arguments.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|