* I did not crop the top and bottom of the image which I should have to
make the final image a nice rectagle. The height for the final image
should not be that of the original images, but rather somethingl like
2*height - accumulator's height.
* I did not test the LivePictures (.ivr) because I could not find the
plugins and did not have time to look further.
The sample image came out nicely. I tried reversing the sequence and
the resulting image is the same.
However, the panorama I took, both using the Kaidan head and the
handheld sequence didn't come out as I expected. The translation from
Lucas-Kanade seems to be off. Due to time constraint, I do not have an
opportunity to solve the problem.
* If my implemention of the code is correct, my guess would be that I
need a different t for Lucas-Kanade.
* Movement in the images for the Kaidan head sequence (view of Red
Square) may also be factor. The similarilty in color may also be a
weakness for precise calculation.
* In the handheld sequence, the nodal point offset may be a factor. The
difference in reflection from the windows between images may also be a
factor.