Crowds in Context: Evaluating the Perceptual Plausibility of Pedestrian Orientations

Abstract

We describe a work-in-progress evaluating the plausibility of pedestrian orientations. While many studies have focused on creating accurate or fast crowd simulation models for populating virtual cities or other environments, little is known about how humans perceive the characteristics of generated scenes. Our initial study, reported here, consists of an evaluation based on static imagery reconstructed from annotated photographs, where the orientations of individuals have been modified. An important focus in our research is the consideration of the effects of the context of the scene on the evaluation, in terms of nearby individuals, objects and the constraints of the walking zone. This work could prove significant for improving and informing the creation of computer graphics pedestrian models. Our particular aim is to inform level-of-detail models