Keynote Speech‎ > ‎

Avideh Zakhor


Avideh Zakhor 

(University of California, Berkeley)

Topic: Image based localization of mobile devices: outdoors and indoors

Pose estimation of mobile devices is useful for a wide variety of applications, including augmented reality and geo-tagging. Even though most of today’s cell phones are equipped with sensors such as GPS, accelerometers, and gyros, the pose estimated via these is often inaccurate. This is particularly true in urban environments where tall buildings block satellite view for GPS, and distortions in Earth’s magnetic field from power lines adversely affect compass readings.  Furthermore in indoor environments, GPS signal is not readily available. In this talk, I describe an approach to image based localization for both indoors and outdoors. This is motivated by the fact that most of today’s cell phones are equipped with cameras whose imagery can be matched against an image database for localization purposes. Our approach consists of two steps. The first step, based on existing work, matches the query image from the cell phone against the image database in order to retrieve a database image of the same scene. The second step recovers rotation and translation via matching point feature correspondences between the query and database images. We characterize the performance of this approach for an outdoor dataset in Oakland, CA and show that for a query set of 92 images, our image based estimation of position is within 10 meters of ground truth for 91% of queries as compared to 31% for GPS on the cell phone. As for indoor localization, the key challenge is a way to develop indoor geo-tagged image database. We use the human operated backpack system made of sensors which was originally designed to generate textured, 3D models of building interiors to build a geo-tagged image database which can then be used for image based localization of mobile devices.