The importance of eye movements to biological visual systems is obvious. Upto now, however, controlled camera movements have played a relatively insignificant role in computer vision research. Active use of camera movements have been increasingly important in the some recent active computer vision systems.
In this thesis, we examine the problem of binocular pursuit of a moving object with a stereo system without discriminating the target from its background. A novel aspect of the approach taken is the use of controlled camera movements to simplify the visual processing necessary to keep the cameras locked on the target. ZDF is good at separating the target object from distracting surroundings but it does not give us information about matching point. Cepstrum filter can accurately estimate the disparity between left and right camera but it can only provide the local disparity. We use these filters to track a moving object. The central idea is to maintain camera gaze by combining ZDF and Cepstrum filter. A gaze-holding system implemented on a stereo head demonstrates this approach and total performance of binocular tracking is demonstrated by actual tracking experiments with a toy.
