Mapping Planets and Moons In 3D With Stereophotoclinometry

subcomdtaco writes with this snippet from a story in the NYTimes: "Dr. [Robert] Gaskell, with software he developed over a quarter-century of trial and error, can process hundreds of images in a few hours, slap them atop one another electronically like coats of paint and produce a topographical map so detailed that you often need a pair of 3-D glasses to appreciate what he has done. At 63, Dr. Gaskell has become the Captain Cook of space. Dr. Gaskell calls what he does 'stereophotoclinometry.' [PDF] Ideally he needs at least three images of the target landscape, usually taken by an orbiting spacecraft or a probe on a flyby to another destination. Only in rare cases can telescope images provide enough detail. The sun angle must be different for each exposure so each image shows different shadows. By comparing the shadows, the software calculates slopes, which yield the altitudes of target features. The computer solves the equation in three dimensions, producing a patchlike topographical maplet."

Comments

This is new?

[pongo000]

We were doing this sort of thing when I worked for Raytheon in the late 90s, using overlapping satellite imagery from IKONOS to generate DTED (digital terrain elevation data) through some rather complex projections. The advantage of this method over Gaskell's is that there is no dependency on sunlight-based data. I fail to see what is new here...

Re:This is new?

[mikael]

There is also a technique called 'photometric stereo'. That takes three images of a rough surface photographed from the same camera position, but with the light source at three different angles. Assuming a basic Lambert lighting model, it is possible to determine the slopes of individual pixels of the image (gradients across and down the image) by converging intensity values back into gradient angles for each axis of the image. Using some discrete FFT calculations, it is possible to convert these gradients back into a heightmap. Shadows aren't really much help as they don't convey any gradient information, and have to be ignored or interpolated.

The only problem is that the light source angles have to be sufficiently far apart in order to get accuravte readings. But this technique can be applied to anything from microscope slides to satellite images.