The localization of the vehicle and the association of the estimated pose is one of the essential tasks in automated driving. Within urban environment, this task is a challenging one, due to the disturbances that interfere the satellite navigation system signals like reflections or multipath propagation. The disturbances result into an erroneous estimation of the ego pose. The effect and its impact depend on the city structure around the vehicle and therefore local correction signals are not useful. In this thesis, a precise localization system is introduced and investigated. A main goal for the developed system is to combine all information the car could provide by its serial hardware and using this information for a stable and precise localization in challenging surroundings. Combining the signals of the cameras to a joined view on the surrounding and using cars odometry information, the localization problem within urban areas is solvable. With the cameras it is possible to detect and measure the relative position of lane segment markings, arrow markings, pedestrian crossings and stop lines. The detection process is presented and evaluated in detail. The landmark information is used with enhanced map data based on Open Street Map (OSM). Thereby, a landmark based estimation can be established. The GNSS information is used for an initial pose guess, the vehicle odometry for position updates and finally the detected landmarks for pose corrections. All information is aggregated within a particle filter for Bayes tracking. It is ensured, that the probability dense function from the particles is a good representation of the actual pose and its probability. The estimation from normal distributed processes is enhanced to a multimodal method. Thereby, the particle filter can demonstrate its benefits. The characteristic of the landmark measurement system is presented and it is shown, how outliers can be identified. Furthermore, the advantage of using multiple cameras in order to improve system availability is presented. A method to associate current measurements with map data by cost function is shown. Additionally it is shown, how the typical resampling method is enhanced, that it supports the multimodal shape of the probability density in best possible way and to optimize it in the state space. It is presented in detail why typical mechanisms to evaluate the particles sets state are not sufficient for urban environment and how to solve this issue. The potential performance of the system is evaluated in test drives in real urban environment, including dense development, roundabouts and tunnels.
