Pose computation based on Correspondence between 2D GIS and Street Images

This article presents a new approach to localization, especially relevant for use in dense urban environments. It is based on a correspondence between image data from pictures taken in the streets, and vector data from a GIS (Geographic Information System). First, a partial geometry of the outdoor scene is computed from the images, for instance angles and length ratios between visible facades. Secondly, this information is transformed into a GIS query, which results in a layer of the possible poses. Then, additional queries allow us to find only one pose, the true one. This can be considered as an alternative to GPS localization, since GPS may perform poorly in a dense urban environment. Moreover, it provides some information on the absolute orientation of the camera, which can be used for instance in Augmented Reality applications. This work is in fact one important step in the creation of a GIS-connected Augmented Reality application, which will allow on site interaction with GIS data.