Orthorectification#
Work in progress
Placeholder content. Being rewritten with figures.
Note
ASP calls this step “mapprojection” in its toolchain (the binary is mapproject, the bundle-adjust flag is --mapproj-dem). This guide uses “orthorectification” — the more standard photogrammetry term — for the concept while keeping ASP’s tool names verbatim.
Resampling each input image onto a reference DEM grid before stereo turns a wide-search-range matching problem into a small-disparity one.
The intuition#
After orthorectification, residual disparity is dominated by error in the reference DEM (small) rather than satellite geometry (large), making stereo matching far easier.
When you can skip it#
Very flat terrain, missing reference DEMs, or quick first passes can use ASP’s --alignment-method affineepipolar instead.
What reference DEM to use#
For Earth, Copernicus GLO-30 (free, on AWS Open Data) is the default; planetary bodies have their own canonical DEMs (MOLA, LOLA, etc.).
The two-pass trick#
Run a coarse first stereo pass on raw imagery, downsample its DEM, orthorectify against that DEM, and re-run stereo — your own DEM is more locally accurate than a global reference.