Computer models of saliency alone fail to predict subjective visual attention to landmarks during observed navigation

Demet Yesiltepe, Ayse Ozbil Torun, Antoine Coutrot, Michael Hornberger, Hugo Spiers, Ruth Conroy Dalton

Research output: Contribution to journalArticle

Abstract

This study aimed to understand whether or not computer models of saliency could explain landmark saliency. An online survey was conducted and participants were asked to watch videos from a spatial navigation video game (Sea Hero Quest). Participants were asked to pay attention to the environments within which the boat was moving and to rate the perceived saliency of each landmark. In addition, state-of-the-art computer saliency models were used to objectively quantify landmark saliency. No significant relationship was found between objective and subjective saliency measures. This indicates that during passive observation of an environment while being navigated, current automated models of saliency fail to predict subjective reports of visual attention to landmarks.

Original languageEnglish
Pages (from-to)39-66
Number of pages28
JournalSpatial Cognition and Computation
Volume21
Issue number1
Early online date12 Oct 2020
DOIs
Publication statusPublished - 2021

Keywords

  • Landmarks
  • object recognition
  • saliency
  • spatial knowledge
  • virtual environments

Cite this