Assessing the perceived realism of agent crowd behaviour within virtual urban environments using psychophysics

  • Stuart O'Connor

    Student thesis: Doctoral ThesisDoctor of Philosophy

    Abstract

    Inhabited virtual environments feature in a growing number of graphical applications. Simulated crowds are employed for different purposes; ranging from evaluation of evacuation procedures to driving interactable elements in video games. For many applications, it is important that the displayed crowd behaviour is perceptually plausible to the intended viewers. Crowd behaviour is inherently in flux, often depending upon many different variables such as location, situation and crowd composition. Researchers have, for a long time, attempted to understand and reason about crowd behaviour, going back as far as famous psychologists such as Gustave Le Bon and Sigmund Freud who applied theories of mob psychology with varying results. Since then, various other methods have been tried, from artificial intelligence to simple heuristics, for crowd simulation. Even though the research into methods for simulating crowds has a long history, evaluating such simulations has received less attention and, as this thesis will show, increased complexity and high-fidelity recreation of recorded behaviours does not guarantee improvement in the plausibility for a human observer. Actual crowd data is not always perceived more real than simulation, making it difficult to identify gold standards, or a ground truth. This thesis presents new work on the use of psychophysics for perceptual evaluation of crowd simulation in order to develop methods and metrics for tailoring crowd behaviour for target applications. Psychophysics itself is branch of psychology dedicated to studying the relationship between a given stimuli and how it is perceived. A three-stage methodology of analysis, synthesis and perception is employed in which crowd data is gathered from the analysis of real instances of crowd behaviour and then used to synthesise behavioural features for simulation before being perceptually evaluated using psychophysics. Perceptual thresholds are calculated based on the psychometric function and key configurations are identified that appear the most perceptually plausible to human viewers. The method is shown to be useful for the initial application and it is expected that it will be applicable to a wide range of simulation problems in which human perception and acceptance is the ultimate measure of success.
    Date of Award2016
    Original languageEnglish
    Awarding Institution
    • Coventry University
    SupervisorChristopher Peters (Supervisor), Fotis Liarokapis (Supervisor), Chrisina Jayne (Supervisor) & James Shuttleworth (Supervisor)

    Cite this

    '