Adding dynamic tracking

We now discuss efforts underway to extend this technology by adding dynamic head tracking capability. The head tracker should provide the location and orientation of the head. For simplicity, we will only consider situations where the head is upright and in the horizontal plane of the speakers. There are two possible uses for the tracking information:

• Update the parameters of the transaural filter based on the tracked head position and orientation. Thus, as the listener moves about, the transaural filter parameters are adjusted to move the ``sweet spot'' along with the listener.
• Update the positions of sound sources according to the tracked head position, so that the auditory scene remains fixed as the listener moves, rather than moving with the listener.

As discussed earlier, the ALIVE project uses video cameras to track people using the system. The Pfinder program can track various features of the human body, including the head and hands [4]. With a single camera viewing a standing person, the distance between the camera and the person is calculated by finding the position of the feet relative to the bottom of the image. Pfinder assumes the person is a vertical plane, and thus the head and hands are assumed to be equidistant from the camera. Features in the plane of the person are determined from the feature positions in the 2-D video image and the calculated distance of the person. Another approach uses stereo cameras and two Pfinder programs to calculate distances to objects by considering parallax. Neither of these systems are currently able to reliably estimate the orientation of the head, because the facial features are too small. However, orientation of the head can be estimated from a closeup view of the face. This is accomplished by obtaining templates for the eyes, nose and mouth, recovering these feature positions via normalized correlation, and assuming an elliptical head shape [8].