Automated Highway Median Landscape Design by Generative Adversarial Networks

Abstract: Highway landscaping performs both aesthetic and functional roles in improving environmental quality and enhancing safety. Studies have indicated that adhering to technical criteria in the design of highway medians leads to a significant reduction in accidents, injuries, and fatalities compared to designs that do not comply. However, in landscaping practice, factors like obstructed visibility at highway intersections due to planting are often overlooked, posing risks to road safety. In this study, we present an automated workflow for the landscape design of highway medians. Our approach leverages a Generative Adversarial Networks (GAN)-based model trained on standard compliant landscape design plans with diverse highway medians geometries. The dataset was sourced from online repositories and subsequently subjected to preprocessing procedures involving paired image-to-image translations, and the conditional GAN architecture Pix2Pix was leveraged. The design output was evaluated using both quantitative metrics and qualitative assessments, with a specific focus on safety considerations. Our study contributes to the implementation of the GAN model in the domain of highway median landscape design, demonstrating the potential for automated quick design ability while adhering to safety standards.