For many applications it is not necessary to know the true location of
the user's face with respect to the world around it, rather it is sufficient
to know the relative position of their face with respect to the image or
its location in a previous image.
There is an initial calibration step where the user's face is located either
automatically (cf. Face Finding,
SegCam, and
Face Recognition
)
or by asking the user to identify it. The appearance of the center portion of the
user's face is then used to initialize a template for subsequent matching, and
the location of their face is used to define an initial search region.
When the system starts running, the template
is convolved through the search region
in the current image. The location where the template best matches
the current image data is taken to be the current face location.
The template and search region are updated based on previous search results.
The general computer vision problem consist of finding the face, determining its orientation, and tracking it accurately over time. The example above is just one way this can be done. After this some version of this process is implemented, the orientation and motion of the user's face can be used for a wide variety of purposes. One of the simplest applications is direct control of cursor position for things like magnifying a particular part of a page. Yet, even for this we must take care to craft suitable transfer functions to convert the user's raw motion into usable control signals. We can also interpret the extracted information at a higher semantic level to control more advanced functions such as document navigation, or to allow the user to dismiss dialog boxes with nods and shakes of their head. |
| Contact: Rick Kjeldsen | Last updated: 6/12/02 | ||
 |
 |
 |
 |