David Forsyth

Ph.D. Oxford University, 1989

Research Statement

Understanding Human Activity

Finding and tracking people: Much research on this topic sees it as a problem in probabilistic inference. I think the significance of the probability and algorithmic aspects of this problem is much overrated; the major difficulties are vision problems.  One has to reason about what things look like, reason about appropriate feature representations, then determine how to find instances of the people in each frame. 

Animating people well requires an understanding of how people move and how to describe what they're doing (so you can compel the animationto do what you want).  Both of these are valuable to the vision community.

Labeling human activity: One of the great problems in computer vision is to say what people are doing from a picture or a video of them doing it. There are numerous applications, ranging from building models of how people behave to advance architectural design, to surveillance. This problem is very hard indeed, for several reasons.   We don't really have a clean vocabulary for what people are doing, particularly for everyday activity.  Often, the interesting stuff is very rare indeed, and people just walk. People can do a lot of different things.  Finally, what they're doing looks different when seen from different angles.

Object recognition is a topic of ongoing interest to me. An attraction is that we don't really know what it means to be doing object recognition, and the collective view of what the problem is has changed rather sharply on several occasions. One should use image cues to identify image regions as containing objects belonging to categories, but what are categories? One source of information is language, and we have looked at the way people annotate images as a cue to what is depicted there.