A three-level motion texture for human motion modeling

Tianyu Huang*, Fengxia Li, Shouyi Zhan

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

A three-level motion texture is proposed to model complex human motion that is statistically similar to the original motion data. The three-level structure, namely moton index, moton and moton distribution, is defined to synthesize motions. To describe the continuous and non-linear dynamics of human motion, the motion texture is modeled by a Non-Stationary Switching Linear Dynamic System (NS-SLDS), which improves the Switching Linear Dynamic System (SLDS) by non-stationary functions. A BSK-tree (Binary Key-pose Splitting Tree) retrieval method applied in motons supplies the ability to access data in frame-level. Thus the motion texture can be manipulated at three different levels, by retrieving key-frame in specific moton, by changing the details of a specific motion at the moton-level and by designing a new choreography at the distribution-level. In motion synthesis experiments, the proposed approach was proved flexible and effective.

Original languageEnglish
Title of host publication2006 IEEE/RSJ International Conference on Intelligent Robots and Systems, IROS 2006
Pages2270-2275
Number of pages6
DOIs
Publication statusPublished - 2006
Event2006 IEEE/RSJ International Conference on Intelligent Robots and Systems, IROS 2006 - Beijing, China
Duration: 9 Oct 200615 Oct 2006

Publication series

NameIEEE International Conference on Intelligent Robots and Systems

Conference

Conference2006 IEEE/RSJ International Conference on Intelligent Robots and Systems, IROS 2006
Country/TerritoryChina
CityBeijing
Period9/10/0615/10/06

Keywords

  • KBS-tree
  • Motion texture
  • Moton
  • NS-SLDS

Fingerprint

Dive into the research topics of 'A three-level motion texture for human motion modeling'. Together they form a unique fingerprint.

Cite this