Virtual City Environments (VCEs) and Mirror Worlds can be a useful resource for communities such as the local government, researchers and the general public to collaborate on tasks like town planning, threat assessment, commerce and research. There are open standards like Extensible 3D (X3D, which represents 3D graphics) and CityGML (a Geography Markup Language to manage 3D building data). These standards are royalty-free and used to create, manage, share and portray such environments. However, there are critical challenges to delivering such complex and detailed Mirror Worlds in real-time.
In this work, we focus on runtime data structures and performance for Level-of-Detail (LOD) management and real-time portrayal. We begin with a VCE defined in existing semantic models such as the CityGML specification. We implement and evaluate a novel X3D-based Level-of-Detail technique called ProxyPrismLOD, which leverages the CityGML standard of a 4-step LOD hierarchy. For switching between different models of the same object at near ranges, our LOD technique uses a custom shape we call a ProxyPrism to optimally encapsulate irregularly and asymmetrically shaped building models.
First, we ran a user study to understand the visual dynamics of range-based LOD switching. Specifically, we evaluated several scaling factors for an exponential range cutoff function. The function is based on the model's size as well as the environment density. In this experiment, participants rated "visual granularity" and "distraction" levels of the LOD technique over two Software Field-of-View (sFOV) conditions. A scaling factor of Beta = 3 was determined. Second, we ran a series of simulations to study the performance benefits of ProxyPrismLOD technique over the basic range-based LOD. We observed performance benefits up to 7.46% in terms of overall Frames-per-Seconds (FPS) on the models we tested.
