Computer vision has been an active research area for the past 30 years. Finding tractable mathematical formulations of many tasks humans take for granted has proven hard and getting the algorithms working in the real world have proven harder.
We believe much of the problem stems from difficulty in aliging images of different view-points, a fundamental step needed to fuse information across multiple images.
This website documents our attempts to fix this long standing problem and apply it to the SLAM and Structure-from-Motion problems.
News: Recently, we received funding to improve correspondence estimation. Our thanks to the Defence Science Organization. Hopefully things will work better!