My work involves writing software to identify classical ballet steps from XYZ tracking data, using methods that will generalize to other kinds of inputs and other domains of human body motion. Classical ballet is a domain with a rich but finite vocabulary of precise movements, several sophisticated description languages, and experts to provide ground truth identification. Currently I am working with XYZ tracking data, which tracks reflectors attached to the major joints of a dancer's body. Identifying steps is a process of converting the "signal" representation of tracking coordinates into a "symbol" representation of named dance steps; thus, in a sense it is "understanding" the motion. Many researchers at many locations are working on recovering human body pose from video. I plan to leverage off their efforts and ultimately build a system capable of understanding human body motion from video. Possible applications include annotating video, smart safety cameras in places like ski lifts and stairwells, and electronic assistants to Olympic judges.