This paper presents a localization mechanism for mobile augmented reality systems in various places. Recently, variety of image-based tracking methods have been proposed: artificial marker based methods, and natural feature based methods. However, localization done with only one tracking methods is difficult in all situation. Therefore, we propose a system, which enables users to continually track in various situation by dynamically switching the multiple localization methods. Our proposed mechanism consists of clients, a switcher, and servers. The server estimates the camera pose of the client, and the switcher selects the outstanding localization method. Furthermore, we employed real-time mapping to continually estimate the position and orientation even if the camera is apart from the prior knowledge of the environment. After localization, the newly updated mapping result is stored in the server. Thus, we could continually track even if the environment has changed.