A visual attentive model for discovering patterns in eye-tracking data—A proposal in cultural heritage

Abstract

In the Cultural Heritage (CH) context, art galleries and museums employ technology devices to enhance and personalise the museum visit experience. However, the most challenging aspect is to determine what the visitor is interested in. In this work, a novel Visual Attentive Model (VAM) has been proposed that is learned from eye tracking data. In particular, eye-tracking data of adults and children observing five paintings with similar characteristics have been collected. The images are selected by CH experts and are-the three “Ideal Cities” (Urbino, Baltimore and Berlin), the Inlaid chest in the National Gallery of Marche and Wooden panel in the “Studiolo del Duca” with Marche view. These pictures have been recognized by experts as having analogous features thus providing coherent visual stimuli. Our proposed method combines a new coordinates representation from eye sequences by using Geometric Algebra with a deep learning model for automated recognition (to identify, differentiate, or authenticate individuals) of people by the attention focus of distinctive eye movement patterns. The experiments were conducted by comparing five Deep Convolutional Neural Networks (DCNNs), yield high accuracy (more than 80%), demonstrating the effectiveness and suitability of the proposed approach in identifying adults and children as museums’ visitors.