Unsupervised learning of human motion models for person detection

This is the work of the recently graduated Yang Song, with whom I have had the pleasure of collaborating with. You can find out more about it from Yang's_webpage.

Goal: "Wherefore art thou, Romeo?"

Although many people have worked on systems for tracking humans, less work has been done in *finding* people (although there has been a lot of work on finding faces).

Yang has developed unsupervised machine learning methods for detecting people (and labelling body parts) based solely on the patterns of visual motion they generate while performing actions.