It actually depends on the contrast in the image. The lower the contrast, the higher the motion can be. In low contrast images you have flat gradients. So what you do in lucas kanade is that you use the gradient to estimate the motion. More precicely you use first order expansion of the L2 based costfunction. Therefore first order expansion can only estimate good results close to the point you are looking at. So flat gradients mean that the distance can be higher cauze the error is lower.

thnx a million time u saved my life with this video

55:10 the last line should be H not H inverse, should be a typo in the slides.

Hello, does anyone know what is the work at 12:09? the one about tracking with multiple fixed & non-overlapping cameras.

thx a lot

11:15 Any good resources on where to look to go further in MTMC ?!!

What is the limitation for the large motion? ... he said if the frames has large motion we need to use pyramids .... how much this large?

Are you missing a ")" around W(x:p)

Hi, the presentasion link is unavailable. Can you provide the new one?

is that understandable for you?

Is anybody here who has dr Mubarak speech in this presenation? pls!!!

Any tutorials on pyramid implementation ?

Thank you!

Course Homepage: www.crcv.ucf.edu/courses/cap5415-fall-2012/

Sir where can we get the matlab implementation of this Algorithm

@41:25 Today a function can be named a great mathematical discovery and understanding. But in 200 years no one will be educated enough to learn anything, muchless a great function!