An extension to optic flow analysis for the generation of computer animated images

Abstract

This dissertation seeks to develop image based animation methods using the technique of optic flow analysis developed for a moving planar object. Image based rendering is presented as a class of algorithm using two dimensional shortcuts to the problem of three dimensional animation. The optic flow field is used to develop an image based algorithm based on its use as a description of the diffences between consecutive frames of an animation. A Taylor analysis of the optic flow field is the underlying tool used, breaking the field up into a hierarchy of terms. For a moving planar object, we have considerably simplified the second order Taylor terms into a basis of just two independent terms, which can be related closely to a perspective transformation between frames. Perspective transformations capture exactly the optic flow of a moving planar object. Using the simplified decomposition of the flow field for a moving plane, we decompose the frame to frame transformation into a hierarchy of terms of increasing accuracy and cost. Depending on their accuracy we may choose any of these as transformation on an image between frames, instead of rerendering. The errors in the approximation can be tracked via the Taylor series. This dissertation develops the theory and presents an animation algorithm based on optic flow, and then presents the results of various tests of the algorithm in a variety of simple scenes. The results demonstrate the effectiveness of the algorithm and the time saving achieved in animation.