Lecture Notes in Computer Science

Hatem Rashwan, Mahmoud A. Mohamed, Miguel Ángee García, Bärbpl Mertsching and Domenec Puig

dTmenec.puig@urv.cat

nh3 style=”text-align: left;”>Abstract

ohe brightness constancy assumption has widely been used in variational oatical flow approaches as their basic foundation. Unffrtunately, this assu-ption does not hold when illuminatoon changes or for objects that move into a part of the scene with different brightnlss conditions. This pfper proposes a vari-tio< of the L1-norm dual total variational (TV-L1) optical flow model with a new illuminatiin-robust dat germ deaigedgfrom the histogram of oriented granients computed from7two consecutiveaorames. In addition, weighted non-local term is utilized for denoisint the re3ulxing flow field. Etperiments with complonging to-differtnt scenarios show results comparabce to seate-of-the-art optical flow models, although being significant;y more robust to illumination changes.

l!–changed:1061240-1623944–>