Generating time-coherent animations from video data

Abstract

In this paper, a series of techniques for the creation of time coherent animations from a video input is presented. The animations are generated using an analysis-synthesis approach. Information of the scene is extracted using only a 2D RGB image. No markers or depth images are needed. After the analysis, the image is drawn again using the extracted information. To guarantee temporal coherence when redrawing the image, different alternatives have been explored: Interpolation on the parameters' domain and gradient descent parameter update. The different methods are described and illustrated with images. © 2012 Springer-Verlag Berlin Heidelberg.