Gesture Recognition

With Ross Cutler, University of Maryland

	Mpeg videos showing short segments: clapping (633 KB), cymbals demo (470 KB), conducting demo (734 KB), drumming (633 KB).
	Paper on this work

We are developing a real-time, view-based gesture recognition system, intended for an environment for children to interact visually with characters on a monitor.  Optical flow is estimated and segmented into motion blobs.  Gestures are recognized using a rule-based technique based on characteristics of the motion blobs such as relative motion and size.  Parameters of the gesture (e.g., frequency) are then estimated using context-specific techniques.

This image shows a frame from a sequence of Ross clapping, and the associates flow output and blob segmentation:

Here am I, flapping my "wings" - the bird on the screen flaps along with me: