Retargetting Motion to New Characters

1
Retargetting Motion to New Characters

• Michael Gleicher

2
Outline

• Introduction
• Constraints and Objectives
• Examples
• Discussion

3
Motion retargetting is to adapt an animated
motion from one character to another,
independently of how the motion was created.
Example of retargetting

• Figure with different sizes or proportions feet
  skating and wrong hands position.
• Constraints essential to the action hands
  position, hands distance apart while carrying,
  feet on the ground.

• A constrained optimization problem
• preserve high-level properties
• Spacetime constraints Make adjustments based on
  all the requirements over frames
• Using the objective function minimized energy
  consumption.

4
Outline

• Introduction
• Constraints and Objectives
• Examples
• Discussion

5
Spacetime Constraints consider relationships
among multiple frames, make choices based on
other parts of the motion. 
Spacetime Constraints
Unsatisfied Results

• Enforce the laws of physics, biomechanical
  limitations
• Equality and Inequality
• Difficult to encode mathematically
• Richer sets are more accurate but may lead to
  difficulty to solve.

Objective function

• Minimizing the amount of change in the important
  properties

IK solver considers each frame independently
6
Constraints are at particular instants of time
come from restrictions on the character,
the environment, or the motion. 
Define the problem
Modeling parameters

• Original motion m0(t)
• New (retargeted) Motion m(t)
• Difference between the two d(t)
• qti parameters of motion at time ti Define
  Constraints f(qti) constant
• Equality Constraints an angle
• through all the frames.
• Inequality Motion dependent. Example In 5
  certain frame the joint angle a2 70 (active set
  method to solve)
• B-Splines Use control points to model the
  angles.

m0(t) m1(t)
Retargetting the motion to a person with longer
arms

• Objective Minimize?(ai - ai)2 for every frame
  t.
• Using B-Splines with control points every 2, 4 or
  8 frames (uniform spacing), then interpolate
  other frames, so that for 8 frames spacing, the
  total parameters of 40 frames can be reduced from
  (24080) to (25 10), which increase the
  speed.

7
To solve this problem, parameters are modeled
using B-Splines and weighted by a sensitivity
matrix M.
M Sensitivities in the variable
Solving procedure

• Start by scaling motion to match scaled
  character, and translation by finding the
  constraint displacements, interpolating and
  smoothing.
• Solve non-linear constraint problem by solving
  the objective equation g(x)½xMx
• M sensitivity function (weight) that summing all
  the displacement of a point in each frame
• x B-Splines model of all the parameters (angles)

• From a1 to a1 the limb motion is not really
  affected much by the change of the angle, while
  the change from a2 to a2 greatly affect the
  motion (sensitive). Use Matrix M to balance the
  parameters.

This part is illustrated in other two papers
Motion editing with spacetime constraints
(1997) Constraint-based Motion adaption (1996)
8
Outline

• Introduction
• Constraints and Objectives
• Examples
• Discussion

9
Examples on motion capture data from an optical
motion capture system, while Euler angle
representations are used for the joints.
Swing dancing
Walking
L original C only female adapted Rboth
adapted.

• A walk adapted to a figure 60 size of the
  original. The smaller one is forced to use the
  original footplant positions. When the
  displacement keys are too distant, overfitting
  causes the wide swings in the yellow foot traces.
  Proper key spacing (blue) results in a motion
  similar to the original (purple).

• Both two characters are adapted, even if only one
  changes size. Hands of the two must remain
  connected in addition to the footplant
  constraints. In center the female gets lifted by
  the hand-hold when spinning and in left both are
  adapted simultaneously with 11 more parameters
  per key.

10
Outline

• Introduction
• Constraints and Objectives
• Examples
• Discussion

11
The computing of the adaption to retarget motions
to a new character can be solved as a constrained
optimization problem.
Discussion
Morphing and differing

• Morphing Use a different scaling amount and
  time-varying translation on each frame.
• Differing characters with similar dimensions
  making constraints on correspondences between
  original and new features and than use standard
  retargetting again.

• Geometric constraints and a simple objective
  function are used.
• Quality of the resulting motions are influenced
  by the complexity of the constraints and
  objective function. Lack of physics constraints
  can lead to unrealistic situations.
• Improved solvers for the numerical problems and
  techniques to avoid the burden of specification,
  would improve the results for a wider range of
  motions.