### Summary

This paper discusses methods for using a timewarp curve, coordinate alignment curve, and a set of constraint matches to correctly blend together two motions. Together, these are called a registration curve. This paper not only presents how to use registration curves to blend motions together, but discusses how to automatically generate registration curves from motions.

### Problem

It is difficult and time consuming to match corresponding points in similar motion data. Further, generating a smoothly varying timewarp curve to align the motions in time is even more difficult. This limits the use of interpolation as only very similar motions can be interpolated between to create a wider range of motions.

### Solution/Methods

Use of a smooth, cubic curve to timewarp motions or parts of motions that are similar as well as a method using dynamic programming to determine which parts of which motions are similar and generate the appropriate curves.

### Contributions

1. Method for determining how close two motions are that is both time and transformation invariant (atleast in terms of translations and rotations about the vertical axis).

2. Interpolation of a wider range of motions by using a timewarp curve.

3. Automatic determination of whether or not two motions are similar and automatic generation of an appropriate timewarp curve.

4. Automatic generation of a corresponding alignment curve given a timewarp curve.

5. Automatic constraint matching and interpolation.