Authors: Robert Pepperell, Nicole Ruta, Alistair Burleigh, Heddwyn Loudon [Cardiff Metropolitan University]

Abstract: Humans naturally see the world from an egocentric perspective. But important questions remain about how best to represent this perspective in image media. Most computer graphics engines use linear perspective (LP) to render 3D space. But previous studies suggest that images created using LP do not accurately represent the experience of being in front of a space, particularly for wide fields of view (FOV) where apparent distortions of object size and shape can occur. We report two experiments in which we compared different methods of representing the egocentric viewpoint in a real-time 3D gaming environment. One method was the standard LP projection used by 3D graphics engines, and the other used a novel projective geometry developed by the authors that is based on the natural structure of human vision, called Natural Rendering (NR). Our results showed that participants: 1) preferred to play the game in NR at a wider FOV setting (M=142°) compared to LP (M=115°) and 2) were 4.9 times more likely to estimate a target object’s correct distance and 5.12 times more likely to give an accurate distance estimation if they were looking at NR renders compared to the LP ones. We conclude that NR approximates the perceived structure of the egocentric viewpoint more closely than LP, providing a more immersive and natural experience of virtual environments, and that it might have benefits for 3D computer graphics more generally.


DOI: DOI: https://dx.doi.org/10.5244/C.33.349
Comments: Presented at BMVC 2019: Workshop on Applications of Egocentric Vision (EgoApp), Cardiff, UK.
Cite as: Paper (PDF): EgoApp2019_3.pdf