Motion synthesis and editing in low‐dimensional spaces

Abstract

Human motion is difficult to create and manipulate because of the high dimensionality and spatiotemporal nature of human motion data. Recently, the use of large collections of captured motion data has added increased realism in character animation. In order to make the synthesis and analysis of motion data tractable, we present a low-dimensional motion space in which high-dimensional human motion can be effectively visualized, synthesized, edited, parameterized, and interpolated in both spatial and temporal domains. Our system allows users to create and edit the motion of animated characters in several ways: The user can sketch and edit a curve on low-dimensional motion space, directly manipulate the character's pose in three-dimensional object space, or specify key poses to create in-between motions. IntroductionCreating animated characters that move realistically is an important problem in computer graphics. One appealing approach is to collect a large amount of human motion data and analyze those data for constructing a behavior model of animated characters. We expect that this model not only contains many primitive actions available to the characters but also is able to provide assorted variants of each primitive action in a parametric form. It is also expected that a set of actions are provided in a connected way so that transitions from one action to another are allowed. Constructing such a behavior model is quite difficult especially if we want to make use of relatively unstructured motion data for behavior generation.With a relatively small collection of motion data, it is possible to look through the entire motion set thoroughly to manually construct a connected set of character behaviors. Through the manual process, one can have an in-depth understanding of what behaviors are available at any given situation and how behaviors are organized. However, this is unrealistic in practical applications because the input motion set must be large *Correspondence to: Hyun Joon Shin, Division of Digital Media, Ajou University, San 5, Woncheon-dong, Yungtong-Ku, Suwon, Korea. E-mail: joony@ajou.ac.kr Contract/grant sponsors: MIC, Korea; Young investigator's award grant from the KOSEF. enough to accommodate a rich variety of natural human motion. Recently developed methods are able to analyze a large motion set automatically to identify a connected set of distinctive behaviors and even parameterize variants of each behavior. The resulting data structure can be searched in order to create a sequence of motions that allows animated characters to track a target or travel along a sketched path. However, this data structure may be too complex for animators to understand how behaviors are organized. From the animator's point of view, it is very important to have direct and immediate control over the character's motion. Animators want to be able to select a set of appropriate behaviors at a given situation and to produce a carefully crafted motion of the character interactively by precisely adjusting parameters of the beha...

Generalizing wave gestures from sparse examples for real-time character control

“…However, this would be possible by embedding into a latent space with extrapolation capabilities, e.g., non-linear Gaussian process latent variable model (GPLVM) embeddings [Lee and Elgammal 2004;Levine et al 2012] and multi-dimensional scaling [Shin and Lee 2006;Cashman and Hormann 2012]. Our focus is instead on robust input generalization -the output of our algorithm could be used as input to these techniques to drive animation synthesis.…”

SkillVis

“…This high-level information has not been used to visualize the skills of the player in previous research. In this research, we combine the approaches of motion graph [Arikan and Forsyth 2002;Lee et al 2002;Kovar et al 2002;Lau and Kuffner 2005;Kwon and Shin 2005] and dimensionality reduction [Grochow et al 2004;Shin and Lee 2006] to visualize high-level skills information of the athletes for the skill assessments.…”

“…For example, Grochow et al [Grochow et al 2004] proposed a method to project the 3D motions of the human onto a 2D plane, and further reconstruct 3D motions by mapping arbitrary points from the 2D plane back onto 3D joint space. PCA [Shin and Lee 2006] and ISOMAP [Tenenbaum et al ;Shum et al 2010] are proposed to map the motions onto 2D planes. Due to the high variation of human motion, local PCA that considers only a relevant subset of the whole motion database in order to generate a locally linear space is proposed Ho et al 2013].…”

Continuous character control with low-dimensional embeddings

“…Prior unsupervised methods created motion variations using graphical models [Lau et al 2009] and used probabilistic techniques and dimensionality reduction to generate new motions that satisfy user constraints or edits [Grochow et al 2004;Shin and Lee 2006;Chai and Hodgins 2007;Urtasun et al 2008;Min et al 2009;Ikemoto et al 2009;Wei et al 2011]. Since our goal is to animate interactive characters using optimal control, we are interested in models that represent motion with low-dimensional spaces, where optimal control methods are tractable.…”