In our new project (starting in 2007), we intend to apply conformal geometric algebra to motion capture to clean up the math in estimating, tracking and filtering `kinematic structures' (like humans) in video images.

We are currently setting up our environment. We intend to use 12 color cameras (slightly above TV resolution, 60 FPS) to find and track (human) motion in regular video images, without using an explicit model. We will focus on using motion clues to estimate conformal rotors. We would like to extend the conformal model to elegantly handle uncertain geometry, and intend to measure the ground truth by doing marker-based motion capture (which has sub centimeter accuracy) at the same time as recording the regular video images.