Animation synthesis by observation and learning

This paper introduces a method for the animation of things based on the observation of natural phenomena and on the synthesis of their behavioral patterns using machine learning methods. The natural phenomena to be animated is recorded using a video camera, and its characteristics behavior is captured. A data sequence representing the subject behavior is obtained from the captured video. By learning the inherent structure in the feature space of some sample data, the learned model can synthesize a novel data sequence from the existing sequences. The generated sequences of behavioral patterns could differ from every original data sequence but preserve characteristics of the subject behavior. We demonstrate the natural animation synthesis through such behavioral pattern sequences, and produce some realistic animation which depict the subject.