An Optical Flow using SVD for Video based Fluid Animation

Abstract

This paper proposes a novel optical flow method using SVD(Singular Value Decomposition) for extracting flow velocity field included in an input video. SVD calculates principal components of the image sequence, and the intermediate frame between sequential images is approximated. We generate optical flow for fluid movements using a groundbreaking optical flow estimation method which is the combination of Lucas/Kanade and Horn/Schunck to achieve the best of the local and global differential methods. While fluid simulation plays central role of generating fluid animations, our video based flow field generation scheme shows the potential for novel paradigm on physics based animation.