Department of Computer Engineering
S E M I N A R
Retargeting Human-Motion Capture Data to Arbitrary 3D Mesh Characters in a Cartoon Style Setting
Computer Engineering Department
Automatic retargeting of human motion capture data to a character with different morphology is a prominent topic of the character animation field. The problem is particularly arduous when the targeted character is non-humanoid, i.e., omposed of vastly different morphology than a human beings. Our work addresses this problem with a novel method combining spacetime formulation of animation with the as-rigid-as-possible surface deformation approach for the purpose of generating animations of arbitrary 3D mesh models from arbitrary human motion capture data. The developed system accepts any prerecorded human motion capture as input and retargets it via training on a small set of matched key poses to the target mesh that may be humanoid or non-humanoid. Retargeting is carried out in a cartoon style setting extending the shape preserving deformation such that the resulting animation of the character conveys certain cartoon animation features while maintaining the semantic content of the input motion. Further work on establishing the training phase with a statistical mapping employing Shared Gaussian Process Latent Variable Models is underway.
DATE: 12 November, 2012, Monday @ 16:50