Abstract
Highly accurate construction of 3D terrain models from satellite imagery currently requires rigorously surveyed ground control points. Capturing this information is expensive, and is required every time an area is surveyed. We present a novel method that matches points and accurate ground coordinates from processed stereo imagery to new imagery of the same area. This provides virtual ground control, removing the need for repeated ground surveys and reducing the cost of repeated model reconstruction for updates and change detection. Our method provides a fully automatic pipeline that significantly reduces errors in the surface model compared to direct georeferencing based on initial knowledge of satellite geometry.