Analysis and Synthesis of Interactive Two-Character Motions

In this paper, we deal with the problem of synthesizing novel motions performed by a pair of human-like characters while reflecting their interactions. Adopting an example-based paradigm, we address three non-trivial issues embedded in this problem: motion labeling, interaction modeling, and motion coupling. For the first issue, we provide a genetic search scheme for motion segmentation and classification, exploiting the structural and geometric features of motion segments. For the second issue, we propose a coupled movement transition graph to capture the interactions between the two characters and their individual behaviors. Based on a coupled hidden Markov chain, the transition graph is trained with example motions. For the last issue, we propose a scheme for synthesizing a novel sequence of coupled movements guided by the transition graph. We believe that our method can be used for two-character motion synthesis in game production.