Augmenting the 3DBAG dataset with facade textures based on real-world images - A case study in AnywhereXR

By Robin Valkenburg

Abstract
This thesis is a proof-of-concept study for enhancing the visual fidelity of 3D urban models by integrating real-world facade textures derived from Google Street View imagery into 3DBAG building geometries of AnywhereXR, an Extended Reality visualization platform. A Generative Adversarial Method (GAN) is tested to understand if making use of a computational heavy method is worth to create better facades for AnywhereXR but was deemed to inaccurate to continue with. The study focuses on two streets in Wageningen, Bergstraat and Bevrijdingsstraat and tries to improve the current limitations of the XR visualization platform by improving the generic, low-detail facade textures with more realistic pictures from Google Street View. The resulting textures are mapped onto 3D models in Blender and visualized in Unity into the AnywhereXR program. A user survey compared two facade types generic and Google Street View based. Results show that the p value of the paired t test is on every question in the survey below the threshold value of alpha = 0.05. In the future research should focus on automation of the integration of Google Street view derived imagery on the 3Dbag dataset in unity and relief texturing to improve the fidelity of the 3D models used inside AnywhereXR.