The stabilize*.m series of files represent the core code - this is the
function you would want to run.

stabilize4.m is probably the best of the bunch.  stabilize5.m was a
(second) attempt at a full affine transformation between frames, but
it has issues with unrestrained skew building up.  Its registration
between frames also seems to suffer, making me wonder if we either
don't have enough interest points to do a good estimation of the extra
parameters, and/or if i've broken some of the math.  (I actually lean
towards the former, since I was staring at the math for a long
time... but there is that whole skew problem... *shrug*)

code is based on TA's code for 16-720, and mostly Tom's homework
assignment code, with additions by both of us.

Ethan did the work to get the full frame rate video capture (JPEG
compression on the Aibo, taking only the Y channel from the YUV images
the Aibo supplies) Tom did the middle work of getting initial results
and testing various concepts, and then Ethan did some more with
calibrating motion and estimating anticipated flow due to turning
(logged on the Aibo when video was recorded)

Tom Stepleton <tss@cs.cmu.edu>
Ethan Tira-Thompson <ejt@cs.cmu.edu>


