Let's look at our palm and wrist: there are 15+ bones inside, not counting the fingers. When we wave a friend good-bye, or operate a mouse, how do these bones move?
We can't place markers directly on the bones to track their motion. We can't shoot an X-ray movie of the moving joint either -- too much radiation can damage our health; and non-radiation imaging techniques have lower resolution. Instead, a common way to analyze joint motion is by CT-imaging the joint bones in several different positions and registering them across all volume images; if you're not familiar with CT-imaging, think of it as X-rays in 3D.
Unfortunately, existing CT-tracking systems may introduce errors on par with the image sampling-step size. In practice, such errors may compromise a motion-study, by introducing interbone collisions.
We developed a subvoxel-accurate method for tracking bone-motion from sequences of CT scans. Results on human wrist data show average accuracy improvements of 74% over the state-of-the-art technique. In practice, that means our method does not introduce inter-bone collisions; we recover correctly the distance between bones, where soft-tissue like cartilage is located.
What this project buys us is: a) a wrist motion database of unprecedented detail; and b) the ability to model and analyze soft-tissue deformation with motion. [see the paper for details]