Évaluation de systèmes d'aide à la conduite. Génération automatique de vérité terrain augmentée à partir d’un capteur haute résolution et d’une cartographie sémantique et 3D ; Evaluation de fonctions de perception tierces

Autonomous driving is one of the current major technological challenges in the automotive sector. Vehicles are becoming more complex and are integrating new systems relying on key functionalities such as perception. Perception is used in various ways to ensure safer mobility, allowing the main inboard system to understand the environment in which the vehicle evolves. Perception plays a critical role in the proper behavior of an autonomous vehicle. It is necessary to ensure that the embedded perception solutions are effective enough to meet safe driving requirements. However, the evaluation of such solutions remains a complex and little explored task. One of the critical issues is the difficulty of generating and having sufficient reference data to conduct relevant evaluations. The purpose of this thesis is to develop a new validation tool to evaluate the performances and error levels of different perception solutions, while minimizing the manual annotations. With this tool, it will be possible to lead benchmark studies on different solutions based on common criteria. The development of this tool is split into two main parts: the automated production of reference data and the evaluation method of the tested perception solutions.