Seminar of the PhD Program
title
Using Large Motion-Capture-Databases for Analysis and Synthesis of Human Motions
speaker
date & place
Wednesday, 27.01.2010, 16:15 h
Room C 252
Room C 252
abstract
Data driven approaches to synthesize human motion is a common approach in computer graphics and animation. In this talk we will focus on two aspects that had obtained previously little attention but have shown to be extremely useful for synthesis and analysis of human motion data and applications such as motion capturing with few sensors.- In many cases the data of motion capture data bases can be represented quite naturally using multi-linear representations using semantic information of the motions (such as performing actor, stylistic variants, ...)
- For semantically unclassified motions in large motion data
bases we have shown recently that using certain medium
dimensional (15-90 dimensional) feature sets for human motions
the task of identifying locally similar regions becomes
practical, if they are used for exact or approximate
kd-tree-based nearest-neighbor searches. On the basis of
kd-tree-based local neighborhood searches also a fast method
for global similarity searches can be devised. We apply these
methods to several tasks such as
- "numerical and logical similarity searches"
- reconstruction of motions from sparse marker sets, and
- building so called "fat graphs",