Computer Vision has made a lot of progress, but many of
the best algorithms don't do as well against textureless regions
with sharply defined boundaries. This trouble is the inspiration
for the Mondrian approach, which relies entirely on sharply
defined edges and textureless regions.
The very basis of the Mondrian approach is the Mondrian
World--a world consisting entirely of fronto-parallel rectangles
at various depths all moving in arbitrary directions (but
maintaining the same depths). Viewed over times, some ideas
could be formed as to which rectangles are in front of others.
This world is highly restricted and not applicable to the
real world, but this site is devoted to relaxing assumptions
and making the Mondrian approach a reasonable one. A summary
of progress is listed under results.
The two subproblems are Mondrian Motion and Mondrian Stereo.
Mondrain Motion involves one camera watching a moving scene,
which allows for movement in any direction. Mondrian stereo
involves different views of the same scene, all in a horizontal
line. It ties velocity indirectly proportional to distance.