Date of Award

2015

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Graduate Group

Computer and Information Science

First Advisor

Norman I. Badler

Abstract

For interactive 3D applications and games, the ability to edit, combine, and adapt motion capture is an important technique for reducing labor, project complexity, and storage overhead. Stylistic motion capture, in particular, has unique challenges since editing can alter the original intent of the actor -- sometimes with poor consequences as viewers can pick up subtle differences in body language. Thus, this thesis aims to enhance motion capture techniques, particularly when it is important that the stylistic intent of a motion is not unintentionally altered, when we might need to quantify the stylistic nature of the motion, and when better tools can help naive users works with large stylistic motion sets. In our first experiments, we investigate how changes introduced by motion editing might alter the emotional content of a motion. We find that emotions were mostly conveyed through the upper body, that the perceived intensity of an emotion can be reduced by blending with a neutral motion, and that posture changes can alter the perceived emotion but subtle changes in dynamics only alter the intensity. In our next experiments, we investigate whether fundamental numerical differences exist between neutral and stylistic motions of a single motion category and find that neutral walks had significantly lower torques than stylistic motions. Next, we model the variation over our stylistic walks using PCA by decoupling each walking example motion into an average and stylistic part such that only the stylistic part is transformed to PCA space. Although our use of PCA for motion modeling is not novel, our decision to decouple motions is a simple enhancement that makes the use of the PCA model produce consistently good output for blending, clustering, searching, and browsing. In our final experiments, we investigate whether using a minimap, created from our PCA model, to browse stylistic motions helps non-expert users search and edit styles. We find that 82\% of participants prefer the map to a label-based interface and that when not under time pressure, users performed equally well with the map, although using fewer mouse clicks and producing more varied results.

Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS