Perception graph for representing visuospatial behavior in virtual environments: A case study for Daejeon City

Abstract

In urban environments, spatial perception is primarily influenced by visual input. Virtual reality (VR) has significantly enhanced our ability to gather comprehensive data on our perception of these environments. However, there is a gap in leveraging these data for the computational analysis of visuospatial perception and its application in spatial contexts. This study introduces a computational model for quantifying visuospatial perception using VR viewing data. We developed a VR city testbed and created a data model to manage geospatial and viewing data. We collected and employed the data to construct perception graphs that represent visuospatial perception. We also computed cell centralities and correlated them with other indicators of visuospatial perception. Using graph neural networks (GNNs), our model identified distinct communities within the VR city. The perception graphs represented spatial connections as perceived by viewers moving in the virtual environment. A significant correlation emerged between the computed centralities and the focus of visual attention, indicating how visual experiences differ depending on various human and spatial factors. This model marks a significant advancement in harnessing viewing data for spatial cognition and offers nuanced and quantitative representations of visuospatial experiences in VR. Furthermore, it extends the application of cognitive graphs to architectural and urban contexts, thereby potentially assisting architects and urban planners in identifying key focal points. By serving as a reference for community detection, this model has the potential to support effective geospatial data management for digital twins. In summary, the model clarifies the context of VR visuospatial experiences and expands the use of cognitive graphs in architecture, including the enhanced detection of communities within a virtual city.