Motion Capture Technologies
- Mechanical
- Magnetic / Ultrasonic
- Inertial
- Passive Optical
- resolution, framerate tradeoffs
- Active Optical
- Vision
Passive Optical Processing
- Image issues (localization in 2D)
- correspondance and matching (2D-2D –> 3D)
- redunandant markers vs. fewer markers
- occlusions and outliers
- “tracking” (which 3D point is what body part – marker ID)
- gap filling
- skeletal solving
- markers –> skeletons
- issues: not an exact model, deal with noise, missing data, …
- coherence issues
- complicated IK problem
- “traditional” pipeline
- track/gap fill/denoise
- skeletal solve
- work with skeletons
- skeleton as an accurate reconstruction of what the performer did
- map skeleton to character
- alternative pipeline
- track / gap / solve
- manipulate marker data
- drive characters from marker data
- who cares what the performer did – the character is different
Power of Mocap
- anything you can get the actor to do
- directors are good at working with actors
- good actors are out there
- easier to explain to an actor what you want than to try to understand it well enough to model it mathematically
- not just “what makes a walk a walk”
- like this character / person
- in this mood
- in this take
- but that specificness is also the curse:
- you get what you get – details and all
- end of day story
What is the motion data?
- sampled at “sample rate” (30hz, 60hz, 120hz, …)
- separate measurement of pose at each time step
- marker data – or skeleton data (angles)
- usually, assume skeleton is fixed over motion
- densely sampled signals
VS. Keyframe data
- standard skeleton (not specialized rig)
- dense vs. sparse
- imperfect (not aligned) – vs. made to work with keyframes
- staggered poses
- keyframe reduction
- not just trying to do spline fitting (but most work does that)
- fitting splines creates more compact representations – not necessarily convenient keyframes
