It is possible to geolocate an outdoor camera using natural scene variations, even when no recognizable features are visible. (left) Example images from three of the cameras from the AMOS dataset. (right) Correlation maps with satellite imagery; a measure of the temporal similarity of the camera variations to satellite pixel variations is color coded in red. The cross shows the maximum correlation point, the star shows the known GPS coordinate.

Overview

A key problem in widely distributed camera networks is geolocating the cameras. In this work we consider three scenarios for camera localization: localizing a camera in an unknown environment using the diurnal cycle, localizing a camera using weather variations by finding correlations with satellite imagery, and adding a new camera in a region with many other cameras.  We find that simple summary statistics (the time course of principal component coefficients) are sufficient to geolocate cameras without determining correspondences between cameras or explicitly reasoning about weather in the scene.

Related Publications

Our work:

• Webcam Geo-localization using Aggregate Light Levels: a formalization of the work in Geolocating Static Cameras
• Geolocating Static Cameras: localization using PCA coefficients of different slices of the dataset, focusing on the diurnal cycle and weather changes
• Consistent Temporal Variations in Many Outdoor Scenes: makes the observations that PCA coefficients encode for weather and sunlight changes
• Toward fully automatic geo-location and geo-orientation of static outdoor cameras: uses the appearance of the sky dome to obtain rough estimates of the camera orientation

Work by others:

• What Do the Sun and the Sky Tell Us About the Camera? : estimation of camera location and orientation. This work builds upon our work in using models from the graphics community to understand images from webcams.
• IM2GPS : geolocating a single image using only visual cues
• What do Color Changes Reveal about an Outdoor Scene?
  

People

The following people have contributed to this project:

• Nathan Jacobs
• Robert Pless
• Kylia Miskell (Washington University in St. Louis, grad)
• Scott Satkin (Washington University in St. Louis, undergrad, to CMU)
• Nathaniel Roman (Washington University in St. Louis, undergrad, to Facebook)
• Richard Speyer (Washington University in St. Louis, undergrad, to Microsoft)

Acknowledgments

Early work on this project was supported under Robert Pless' NSF CAREER award: NSF IIS 0546383: CAREER: Passive Vision, What Can Be Learned by a Stationary Observer. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.