Eva Schiffer

Snap Together Motion
by Mike Gleicher et al. 2003 Symposium on Interactive 3D Graphics. 

Summary:

This paper describes a system for creating very compact motion graphs which represent all possible character poses and transitions in a strongly connected way and using them to power 3D characters movement in real time from user input. 

Problem:

This paper essentially tackles the problem of creating traditional move trees in a semi-automated way from motion capture data. 

Method:

This paper presents a semi-automated, user driven method for selecting match sets from motion capture data. The system is able to do various automated things, such as selecting likely match sets for the user to chose from, or detecting outlying poses which could lead to a degeneration of motion, in order to help the user find the match sets that will work best for the given data. Once the match sets are chosen, a simplified motion graph is formed from the input data by creating common poses for each match set and then generating transitions to and from these common poses automatically by blending the input data. 

Key Ideas:

It is possible to create useful move trees from motion capture data quickly and with relatively little artistic skill. 
The semi-automated process allows move trees to be more accessible to applications which could no previously afford expensive artists to create them by hand.

Contributions:

The semi-automated system presented in this paper allows move trees to be created very quickly and easily as long as you have motion capture data which represents the sort of motions that you want your character to be able to perform. This paper also presents a way to create C1 displacement maps, which allow for smoother transitions between the common poses.

Questions:

The system that they described for editing the match sets to create the move trees seemed very interesting to me. I was curious if this system was really as intuitive and fast as they claimed.