In recent years, by the spread of sharing services of images and videos, we have come to be able to acquire landscape images and vehicle mounted videos of several cities in the world from the Internet. This information will be useful for scene understanding, the three-dimensional model generation of the city and the frequent update of the map, but may provide the wrong information when the data has no correct geographical information. On the other hand, many researchers try to take the information that a human being perceives out of the information about the scene and study on labeling of the scene, but there are few successful methods to identify images taken from plural cities. Therefore, we aim for the global localization of images which are taken in various cities in the world using only the visual information. To this end, we learned the information of each city from a Google street view image and constructed the city identi_cation database, then estimated the city where an inputted image taken from. In ordert to con_rm the effectiveness of proposed methods, we proved it using images acquired from 15 cities and obtained several knowledge.
View full abstract