Evaluation of the accuracy of human eye movement system identification using step test signals

Abstract

This study investigates nonlinear dynamic system identification methods to simulate the human eye movement system (EMS),
focusing on the accurate representation of transient characteristics derived from step test signals. Integral nonlinear models were
applied to capture the nonlinear dynamics and inertial properties of the EMS. Experimental "input-output" data were collected using
advanced eye-tracking technology, enabling the identification of multidimensional transient characteristics (MTCs) that describe the
EMS’s dynamic behavior in response to visual stimuli. The research utilized approximation and compensation methods to develop
models based on integro-power series (IPS), while the least squares method (LSM) was applied to construct integro-power
polynomial (IPP) models. The compensation method, while less computationally demanding, showed lower accuracy, making it less
applicable for tasks requiring high precision. Third-order models exhibited instability in their transient characteristics, limiting their
practical use. Second-order models, specifically quadratic IPP models developed with LSM, proved to be the most accurate and
computationally efficient. These models provided precise and consistent representations of EMS dynamics, with error rates
significantly reduced when using three test signal responses instead of two. This emphasizes the importance of sufficient data in
improving model reliability. The findings highlight the suitability of the quadratic IPP model refined with LSM for further
investigations. This model offers a robust basis for advancing research into personalized psychophysiological condition assessment
through the development of classifiers. Its accuracy and stability make it a valuable tool for exploring state classification
methodologies in healthcare, cognitive science, and related domains requiring precise dynamic system simulation.