Eva Schiffer

Flexible Automatic Motion Blending with Registration Curves
by Lucas Kovar and Michael Gleicher. 2003 Symposium on Computer Animation. 

Summary:

This paper presents the registration curve data structure which encodes information about the best identifiable blends between motions. 

Problem:

The problem this paper tackled is how to encode information about the similarity and difference between motions such that blends of the motions can be calculated easily and well. The sub-problem of how to identify which points along motions are similar is also discussed. 

Method:

This paper phrases the similarity problem for motions as a dynamic programming problem based on finding the minimal paths in a chart of the optimal squared distances between point clouds representing a frame of the motions at a particular time. Once the minimal error paths are found, they define a time-warp curve for the two motions involved. Using this time-warp curve, the two motions can be blended fairly simply. Existing constraints are matched before blending in an iterative process.

Key Ideas:

Finding the time-warp curves of similar motions can be tackled quickly as a dynamic programming problem if we limit how the time-warp curves are allowed to travel around in the minimum squared error chart. 
Using existing minimum error paths, it is possible to match constraints in a simple iterative process.
Time-warp curves allow for powerful control of motion blends.

Contributions:

The system for creating time-warp curves using dynamic programming is the major contribution of this paper. 

Questions:

I am curious to know how well this system handles matching constraints. Are very noticeable artifacts often introduced?