Captivated by motion capture
Computers can animate a person jumping rope or skipping down the street in several ways. It's probably easiest to use a shortcut called motion capture, which digitizes real-life motions by recording the position of key points like the hands, elbows and knees. Wearing special reflectors, the actor performs before a set of cameras that record the position of each reflector in three dimensions. These positions are entered in a database and later manipulated by software.
Very cool animation. 928K
©1996, Computer Graphics Lab, EPFL - LIG, Lausanne, Switzerland.
A Real-Time Anatomical Converter for Human Motion Capture gives real-time conversion of magnetic sensors measurements into human anatomical rotations.
Motion capture is quick, fast, and effective. But what if one of the characters is too big? Things can get strange -- people can crash into each other or rise off the floor, as you can see from our sample video. In computerese, they violate "constraints."
1. This original shows professional swing dancers in action.
2. If the female partner shrinks, she'd need some serious platform shoes using an older approach to motion capture!
3. By mathematically adjusting the position of both dancers, things look lifelike once again.
Download the movies here.
ORIGINAL, 1.2 MB
UNADAPTED RE-
ADAPT BOTH MOTIONS, 1.1 MB
OR
ALL THREE with sound! 2.7MB Courtesy of Michael Gleicher, University of Wisconsin-Madison department of computer science.
Some people have tried to fix the problem by altering one frame at a time, but that produces jerky, unnatural motion.
To avoid the need to reshoot the scene using a different size actor, and to shrink animation costs, Michael Gleicher, a University of Wisconsin-Madison computer scientist, is developing software tricks that alter sizes without making animated people float over the floor or act jerky. When the smaller partner shrank in his sample, he directed the software to "Give me a motion that's most similar to the original," while obeying two ground rules -- the partners had to hold hands, and their feet had to touch the floor (none of those 12-inch platform shoes, puleezze!)
The math gets hairy, he says, since a smooth animation requires that adjacent scenes relate to each other. Since the two dancers carried a total of 40 motion points, animating 247 frames required solving an equation with 40 * 247, or 9880, variables.
Ouch!
Although motion capture may seem like a shortcut compared to keyframing and simulation, Gleicher says "That's a fallacy. It moves the problem, requires a different set of talents. If you want an [conventional] animation of a ballet dancer, you need an animator who knows a lot about ballet." To do the same thing with motion capture, "You need a ballet dancer."
Seems like simulation's next, eh?
