Procedural improvement of models and textures of 3D urban scenery: A case study

Human-in-the-loop driving simulation experiments are used in academic and industry studies for a variety of purposes that has been growing since the introduction of autonomous vehicles. The preparation of these studies involves the creation of several miles of streets and road which can be very expensive and time-consuming, while they are often restricted by serious budget limitations and short deadlines. Procedural graphics generation is a broad umbrella term that groups a diversity of methodologies to generate graphics models algorithmically and that has received a strong interest from several industry sectors precisely to lower this creation cost. In this paper, we review different commercial and open-source tools and demonstrate how procedural approaches can be used to create new geotypical models and to put together the set of scenery that is needed for participants to have an immersive driving experience. Finally, we conduct a performance analysis of the generated models under driving conditions, revealing their performance bottlenecks and the potential optimisations to address them. The results show that it is possible to improve the graphical quality of large environments for real-time simulation with limited means, improving the immersive experience by matching the expectations of the participants, which should encourage small teams to develop state-of-the-art graphical models for their virtual environments.