Precise localization for mobile Augmented Reality in large indoor environments without specific tracking infrastructure is challenging. This is especially true for rooms with changing properties, like lighting, seating and carpeting. With these constraints a map for a vision based tracking approach has to be continuously updated. The Parallel Tracking and Mapping (PTAM) algorithm is capable of generating and extending a map while tracking the camera pose in an unknown environment. However, it has originally been designed for small workspace environments and has therefore certain limitations. We have extended and modified the original implementation in order to ensure efficient and robust map generation and tracking in large rooms. Furthermore, we have tested a mobile setup with the system in Festsaal in Vienna's Hofburg, which is close to thousand square meters in size. The user's position and path was tracked while the environment was augmented with virtual objects and the system was successfully tested for robustness and occlusions.