The work describes novel measurement models for the localization of vehicles in urban environments. The system utilizes different sensors which are mounted to a vehicle: A gray scale camera, vehicle motion sensors and a GPS receiver. Moreover, digital maps are used as additional source of information. The innovation of the presented method is in the novel kind of incorporation of the image data delivered by a camera sensor into the process of estimating the vehicle's position. In contrast to state-of-the-art approaches, the models proposed by the author are able to include entire image areas instead of only distinct features limited in terms of space. The approaches use different means of representing the digital map data. On the one hand, data equivalent to standardized map databases can be utilized. On the other hand, aerial images can be used as well.
The presented approaches are evaluated using real-world data. For this purpose, measurements were provided by a test vehicle and an extensive test drive. The results of the algorithm are compared to a ground-truth reference. It is shown that the approach proposed by the author can achieve lane-level accuracy of the position information in urban environments using the given sensors. This enables new kinds of applications, at the same time keeping the costs for such system feasible.