Behavior Learning and Animation Synthesis of Falling Flat Objects
Kohta Aoki*, Osamu Hasegawa**,***, and Hiroshi Nagahashi**
*Interdisciplinary Graduate School of Science and Engineering, Tokyo Institute of Technology
**Imaging Science and Engineering Laboratory, Tokyo Institute of Technology, 4259 Nagatsuta-cho, Midori-ku, Yokohama 226-8503, Japan
***PRESTO, Japan Science and Technology Corp. (JST)
In this paper, we describe an approach to learning patterns from sample data sequences and generating new data sequences through learned models. The target application of this work is the animation of natural phenomena, especially falling behavior of flat objects. The natural object or phenomenon to be animated is recorded using one camera, and its characteristic behavior is captured. Feature vectors are defined as the representation of behavior and are automatically extracted from captured videos. By learning the structure of a set of sample vector sequences, the learned model can generate a novel pattern through the underlying structure. These generated patterns could differ from every original vector sequence but preserve characteristics of subject behavior. We can use such patterns to synthesize natural-looking animation.