Motion Capture

Tracking motion of reference joints.

Joints’ positions $\rightarrow$ joint angles which drive an articulated model/deformable surface.

Technologies

Passive optical systems:

• Retroreflective markers that reflect light generated near camera lens.
• Sampling of only the bright reflective markers
• Pros: higher fps
• Cons: no glossy/reflective materials, occlusion of markers by props/limbs

e.g. Acclaim Motion Analysis

• 240Hz, non-real time
• 3 markers per body part (to achieve 6 dof)
• 2+ cameras for 3D position
• Expensive ($100K)

Magnetic systmes:

• position and orientation calculations via magnetic flux
• 3 orthogonal coils on transmitter and each receiver
• sensor output is 6DOF
• metal in environment can cause interference
• low frequency (30 to 120Hz), few markers
• Less expensive ($40K)

Monkey:

• mechanical puppet with potentiometers measuring flex of joints
• High accuracy and frequency
• But needs animator.

Processing

• Filtering raw data (e.g. Retargeting)
• Inverse Kinematics application

Retargeting

Taking captured motion, applying it to another arbitrary figure without destroying motion.

Subproblems:

• Cyclification ??
• Transitions
• Generalizing motions
• Extracting style
• Scaling

Process

1. Define constraints
2. Apply to a new character
3. Add any translation offsets (obtaining a approximate answer)
4. Solve for constraints from part 1 (non-linear constrained optimization)
   1. e.g. Foot doesn’t penetrate through floor/clip through leg etc.

Motion graph

Selecting a sequence of motions from raw animation data to generate a new animation sequence

Edges: The motion clip (e.g. walking, running, squatting, jumping)

Nodes: Choice points for possible successor edges

Transitions: movements which connect two different animation samples at a node.

To create a transition:

1. Select transition point candidates
2. Elimination of non ideal transition points
3. Blending frames

To extract a sequence of motions: Cast as a graph search problem, with a branch-and-bound method

To synthesise path: estimate actual path $P’$ and compare with desired path $P$, transforming movement until error towards $P$