Lecture 14: Editing and Blending

Motions as Signals

Motion signal intuitions

• frequency content – important
  • high frequencies are important (presence or absence)
  • over-smoothing
• frequency-based noised reducion

Time manipulations

• shifts
• scalings
• warpings
  • how to find the warps?  manually – automatic later
• resampling
• messing with derivatives

Space manipulations

• transform the whole motion
• scale / rotate / translate / …
• align motions

Treating the root seperately

• move root (by high-level control) – then add details
• better to think of as “floor tile” (since hip motion is useful) – global transform to character
• x,z,theta (along floor) is special

Manipulations of the non-root

• per-frame changes?
  • need to be careful! (coherence)
• per-motion changes? (constants)
  • adding constants (tilt head)
  • scaling (emphasize)
• changes change over time
  • add another motion? (linear combination)
    • blending – we’ll come back to it
  • add a displacement function
    • motion warps / motion displacement maps
    • layering
    • temporal locality control (independence of edits and frame rate)
    • posture interpolation at key frames

Displacement Maps

• basic idea
• IK at keyframes
• layering

Constraint based issues

• IK at key frames
• IK per frame
• Spacetime
• PFIK + F

Retargeting

Blending

• Basic idea
• Why does it work?
• manifold of pose space
• manifold of motion space
• “real” interpolation vs. interpolation in parameters
• When doesn’t it work?
• Interpolating physical properties

Making it work

• Capturing Wisely
• Choosing Wisely
• Editing Wisely
• Manual Alignment
• global position
• key time time alignment (manual time warps)
• problem: tedious, only a few points, …

How to decide automatically?

• Distance metric