Simulationsgestützte Landmarkendetektion, Lokalisierung und Modellgenerierung für mobile Systeme

For the meaningful interaction of mobile systems with their environment, a geometric and semantic environment detection and description is fundamental. The quintessence of this dissertation is a concept for model-based environment detection, coupled with object-based localization. The goal is to describe the environment in semantic environmental models and, in conjunction with a simulation system, to represent it as a comprehensive semantic world model. The concept of the present dissertation includes the three main components environment detection, localization and environmental modeling. The recognition of surrounding objects and the assignment of suitable semantics is presented on the basis of the development and implementation of sensor data processing algorithms for different application scenarios. The implemented landmark-based localization makes it possible to locate the detected objects in a common frame of reference so that they can subsequently link additional data sources. The broad spectrum of applications starts with simple scenarios of mobile robotics and extends to the participation of fully autonomous automobiles on the road. With increasing complexity of the field of application, the requirements for the environmental detection and description also increase. In addition, a focus on the concept of situational awareness takes place, starting with the observation of data from individual sensors, through the semantic interpretation of the measurements, up to application-spanning, holistic modeling in the simulation system. The present work extends the currently predominant notion of primarily sensory environment detection and modeling to the essential aspects of semantic recognition, description and management. Semantic modeling similar to human perception of the environment enables the implementation of complex applications that "understand" their environment. Thus, they need to be able to capture, classify and, in some cases, to abstract complex relationships. Eventually, they are able to interact safely and reasonably. The introduction of a formal and modular concept, which interconnects the three domains, creates a uniform and comprehensive solution for various applications. The present work lays the foundation for further developments in the field of collectively usable, multi-application world models and the realized applications show the impressive potential of this research.