This paper introduces an automatic method to detect motion data
that can be successfully blended using registration curves.
Registration curves can be divided into 3 parts.
First, time warping curve to line up the timeline between 2 motions.
Second, coordinate alignment to transform motions so they are
as similar as possible. Third, matching constraints.
The paper describes methods using DTW to align time and use
previous steps to create alignment curves and identifying constraints.
To make this method find more accurate matches, I was wondering
if using the del position of each point cloud instead of its
world position might enable us to better find logically similar
motions with very differ point cloud positions.
eg. very low kick & very high kick