Motion Capture Technologies
Jessica Hodgins
• Animation
• Video Games • Robot Control
Motion Capture
What games use motion capture?
• NBA live
• PGA tour
• NHL hockey
• Legends of Wrestling 2
• Lords of Everquest
• Lord of the Rings
• Tony Hawk Pro Skater
• Batman: Dark Tomorrow
• Grand Theft Auto
• Soul Caliber
Motion Capture
• Track motion of reference points
• Convert to joint angles
• Use angles to drive an articulated model
• Motion paths can then be combined to give greater control
What is Captured?
• What do we need to know?
– X,Y,Z
– Roll, pitch, yaw
Errors cause
– Joints to come apart – Links to grow/shrink – Bad contact points
• Off-line
How to use the data
– Processing by filter, inverse kinematics
– Produce libraries of motion • Choose among them
• Blend between them
• Modify on the fly
• On-line (performance animation)
– Driving character directly based on what user does in real-time. Konami games (for a few dof)
Production Pipeline
What is captured?
• Dynamic motions?
House of Moves
• Scale?
What is captured?
Motion Analysis
What is captured?
• Non-rigid objects?
House of Moves
What is captured?
• Props often cause problems
– Ball in pingpong – Fly fishing
– Sword
• Passive behaviors are hard
– Complicated motion of clothing – Explosions
• Calibration
Technology Issues
• Resolution/range of motion
• Accuracy
– Marker movement – Capture rate
• Occlusion/correspondence
Marker Placement
• Location should move rigidly with limb
• Stay away from bulging muscles
• Shoulders: skeletal motion not closely tied to skin
• Shoulders: not a rigid body chain
Marker Set
• 3Dpoints->jointangles
• Itmatters!
• Sodoassumptionsabouthumankinematics
• Calibration of cameras
Calibration
– Fairly automatic with L-bracket and wand
• Calibration of subject
– Also fairly automatic but not so accurate – What are limb lengths
– What are joint limits?
– Where are markers wrt to joints?
Eric Darnell, co-director of Antz
The main problem with motion capture associated with characters has to do with mass distribution, weight, and exaggeration. He says that it is impossible for a performer to produce the kind of motion exaggeration that a cartoon character needs, and the mass and weight of the performer almost never looks good when applied to a character of different proportions.
Richard Chuang, VP at PDI
The mapping of human motion to a character with non-human proportions doesn’t work, because the most important things you get out motion capture are the weight shifts and the subtleties and that balancing act of the human body. If the proportions change, you throw all that out the door, so you might as well animate it.
Godzilla: Karen Goulekas
The reason that we pulled the plug on using the motion capture was, very simply, because the motion we captured from the human actor could not give us the lizard-like motion we were seeking. The mocap could also not reflect the huge mass of Godzilla either. During our keyframe tests, we found that the Godzilla motion we wanted was one that maintained the sense of huge mass and weight while still moving in a graceful and agile manner. No human actor could give us this result.
• Retargeting
Research Topics
• Marker placement/extraction of skeleton
• Capture of deforming skin rather than pretending it is a rigid chain?
• Constraint satisfaction
• Generalization of data – Interpolation
• Interfaces for controlling human motion
Motion Database
User Specified Path
Jehee Lee, Jinxiang Chai, Paul Reitsma, Jessica Hodgins, Nancy Pollard SIGGRAPH 2002
Find Transitions and Re-sequence
Motion Capture Region Virtual Environment
Sketched path
Obstacles
Find Transitions and Re-sequence
Motion Capture Region Virtual Environment
Unstructured Input Data
Connecting Transitions
Synthetic Transitions
Good transitions taken with high probability
“Video Texture” [Schoedl et al. SIGGRAPH 2000]
– Poses
– Velocities
– Contact state
Search to Find Path
Motion Data for Rough Terrain
Motion Data for Rough Terrain
Comparison to Real Motion
Comparison to Real Motion
What are the tricks?
• Enough data?
– Wash out yaw
– Only locomotion
• Wash out yaw?
Why isn’t this used in industry?
• No control over transitions?
• More dynamic behaviors so automatic transitions would be few?
• Havetoplanforshootanywaysomightas well plan for transitions?
• Notenoughdata?
Next lecture
• Details on transition metrics
• Details on washing out yaw
• Representations for orientation – Euler angles
– Quaternions – Axis/angle
• A bit about editing, generalizing