We have applied techniques from the "structure from motion" literature to the problem of extracting high-resolution 3-D topographic surfaces from unconstrained photographs. This technique requires the selection of corresponding points in three or more photographs. From these corresponding points the unknown camera positions and surface topography are simultaneously estimated.

Shown to the right is a ridge-valley sequence from the Oregon Coast Range, and a portion of the estimated surface topography.

Below is the Matlab code for estimating surface topography from unconstrained photographs. Save each file with the same name as listed below, and see phototop.m for instructions.

   phototop.m
   rotmat.m
   errfuncperspM.m
   errfuncperspS.m
   errfuncperspSmooth.m