Cooperative gaze holding in binocular vision

Vision systems that hold their gaze on a visual target using binocular, maneuverable computer vision hardware are discussed. The benefits of gaze holding are identified, and the role of binocular cues and vergence in implementing a gaze-holding system is addressed. A combination of Smith prediction and optimal signal estimation that allows a system of several simulated interacting gaze-holding controls to act together coherently, despite delays and shared output variables, to produce zero-latency tracking of a moving target is described. It is shown how tracking the object increases its signal by improving its image quality and decreases the signal of competing objects by decreasing their image quality. The implementation of a subset of the gaze-holding capabilities (gross vergence and smooth tracking) on real-time computer vision hardware is described.<>