http://www.smartertechnology.com/c/a/Smarter-Strategies/Cloud-Makes-3D-Models-from-Ariel-Photos/
I have mixed feeling about this.
Theory:
From a number of overlapping photos, common points can be determined, and the viewpoint of each picture (relative to the view points of the other pictures) can be computed straight forwardly, and given the calculations of the view points of each, and identifiying the 3D coordinates is (in the mathematical sense) of points in overlap areas can be assigned 3 D coordinates.
You'd want at LEAST 50% overlap (I.E. every point on the ground appears in at least two pictures.)
Practice:
The accuracy varies, based on the usual "strength of figure" arguments in a new guise. What people get is basicly a point cloud, without a clue as to the accuracy of the individual points, without the viewpoint relationships (and coordinates), and with an assumed level.
Social issues:
The ads are already calling the result a "map", and people will get points to a gazillion digits... most of which won't be significant in any way.
And, I'm told, you can soon have the original pictures drapped over the new 3d surface.
It will, in my opinion, the same issues dealing with some GIS newbies now: "I was given 20 digits, I must have 20 digits of accuracy."
Bright side:
In the right hands, with some identifiable control points, and some software to rotate/scale/fit the cloud to control points, this could be a cheap fast way of getting a 3d model of what the land and structures look like.
