Computing structure from motion is a problem of underdetermination that has been a focus of interest in the study of vision by natural computation researchers. It has long been known that when humans view motion -- even simple movements of a small set of dots -- this motion can be used to provide a powerful sense of 3D structure. This is called the kinetic depth effect. How is this effect possible?
Ullman (1979) considered simple motion displays consisting of dots. He assumed that the motion correspondence problem had been solved for this display. He then adopted another assumption: that the object in motion was rigid. He then developed a proof that it was possible to deduce the 3D structure of the moving object using only three different views, where each view required only four different dots. The proof required that the four points not be coplanar. This work inspired a great deal of later work in the natural computation approach to vision, in which researchers explored natural constraints that could be used to (formally) solve visual problems of underdetermination.
- Ullman, S. (1979). The interpretation of visual motion. Cambridge, MA: MIT Press.
(Added November, 2009)