A space-variant approach to oculomotor control

Vision is a very important issue developing artificial anthropomorphic systems. Starting from traditional knowledge about ocular movements, mainly due to physiological observations, the paper proposes some new ideas on gaze control. It is shown how smooth pursuit and vergence can be considerably simplified by using space-variant vision. The implementation of binocular fusion becomes a key aspect in order to segment a candidate target from the background and perform continuous tracking over time. A set of experiments on real data demonstrates that space-variant processes can be fruitfully integrated in a simple hardware platform.