Synthesis and editing of realistic urban road networks

Abstract

3D modelling of realistic road networks is beneficial to a number of areas including urban planning and traffic management. Simulations carried out in these environments are key to the development and testing of software used to drive autonomous vehicles. The generation of these 3D models is a laborious task with considerable time spent ensuring that the created environments represent the real world accurately. The aim of this project is to design and develop a system to bootstrap this task by procedurally generating road geometry which fuses information from multiple freely-available sources. Blender is a popular open-source 3D modelling software with extensive support for scripting; all functionality available through its GUI is exposed through its Python scripting API. This positions Blender as an ideal candidate for the procedural generation of geometry which can then be further refined within the GUI. Several Python libraries are readily accessible to carry out important tasks for this project such as network data queries and image processing. The main mapping information source for this project is OpenStreetMap which is built by a community of mappers that contribute data about roads and buildings around the world. Since OpenStreetMap does not provide satellite views, satellite images from Google Maps are used and aligned to Blender's procedurally generated road geometry. Image processing is carried out on these colour images to calculate the real widths of roads and approximate the location of greenery. Elevation data from Google's Elevation API is used to further improve accuracy. OSM2Blender, the system developed for this project, produces very accurate results which in some cases improve upon existing solutions like BlenderGIS. Locations covering a variety of urban layouts were chosen to evaluate the generated geometry's quality. Future work includes the procedural addition of buildings and applying materials to increase the realism of the generated results.