In order to achieve high quality tracking performance, the face tracker uses a Canon VC-C1 camera with pan, tilt, and zoom control. There are two major problems with this camera: (1) the camera cannot pan and tilt simultaneously; (2) The response of the camera is much slower compared to the real time sampling rate. We have developed several methods solve these problems. Instead of directly controlling the camera, we use a socket-based server. With the server, client code does not have to deal with complex RS-232 port and client code can ignore the fact that the VC-C1 does not have simultaneous pan, tilt or zoom. If we use a conventional feedback control scheme, we can hardly achieve good performance because of time-delay. To overcome time-delay, we have developed a model-based predictive feedback scheme as shown in Figure 7.