Human point-of-regard tracking using state space and modular neural network models

The presence of saccadic and smooth movements in the eye makes modular neural networks composed of two experts, each individually responsible for saccadic and smooth movements of the eyes, well suited for the tracking of human point-of-regard. To establish a basis for comparison on our data, we also consider a scalar ARMA model and a (vector) state space model. The purpose of this analysis is to build a reasonable model of human eye motion to use in prediction of point-of-regard. The ability to predict the point-of-regard of a human subject has applications in eye-tracking for man-machine interfacing, vigilance detection, and as a tool in cognitive psychology.