• 
• 
• 
• 
• 
• 

Camera Network Calibration

Cameras are distributed throughout a city.  From their images, we determine the 3D locations of the cameras and many of the environmental features they view.  This makes it possible to quickly survey and build 3D models of cities.

Background & Objective:  Computer vision has intensely studied the problem of building 3D models of rooms and plazas from a dense set of images.  We seek to build a 3D model of an entire city from a sparse set of images that might be collected by walking around with a camera.

Technical Discussion:  By examining the vanishing points in images, it is possible to determine camera orientations. By examining environmental features (such as building corners) viewed by two cameras, it is possible to determine the direction from one camera to the other.  The missing information is the relative locations of the cameras and environmental features in 3D.  We developed a closed-form least-squares solution, and computed centimeter-accurate localizations from kilometer-scale camera networks.

Technology Area:  Computer Vision

Modification Date:  September 12, 2007