General Construction of Time-Domain Filters for Orientation Data (Lee and Shin)

Being able to filter orientation data can be incredibly useful. It can
provide a means of smoothing out an animation. Further, it can help in
retargeting an animation to other models, since this essentially
becomes a resampling problem. This paper discusses a method of
constructing coordinate-invariant, time-invariant, and symmetric
filters for orientation data.

Key ideas/Contributions:

1. A method of relating a quaternion signal to its vector counterpart
by associating angular displacement in the quaternion signal with
linear displacement in the vector signal.

2. A filtering algorithm based on this observation, allowing
coordinate-invariant, time-invariant, and symmetric filters to be
constructed and applied to orientation signals.

Questions:

While the mathematics make sense, and the graphs and figures provided
match what I would expect out of this work, I would like to see some
examples of this algorithm in action to get a more intuitive sense of
why and when to use it. It is much easier to see the effects when
watching an animation as opposed to looking at a few still images and
graphs.