MODELING PREDICTIVE TRACKING OF EYE MOVEMENT CONTROL SYSTEM WITH A SIMPLE NEURAL NETWORK

摘要

The rapid adaptation seen in human ocular tracking of repetitive stimuli is both functionally important and difficult to model. A modified back propagation network was trained to predict future eye velocity using error signals from a human subject. It accurately tracked a sinusoid at the training frequency and showed reasonably good ability to track at other frequencies. Eventual incorporation of such an element into existing models of transient pursuit could produce a more broadly useful tracking system for active vision applications.