2:00 PM - 2:15 PM
[034] Automated 3D Urban Site Design Method for Initial Design using Generative Adversarial Network
Keywords:Urban Design, Generative Adversarial Network, Artificial Intelligence
Urban design process requires expensive time and cost, which is conducted by well-trained experts and number of stakeholders. This study suggests artificial intelligence model using Generative Adversarial Network (GAN) for automated urban site design. We extract block images which include land use and Floor Area Ratio (FAR) information of Seoul to train GAN model. Trained model generates new urban layout of given site – Sewoon Sanga, including land use configuration and FAR. The generated urban site in 3D shows visual output which can be used for initial design examination. Our model generates suitable design for target criteria including site area, layout, and FAR based on existing urban context of Seoul. First, our model differentiates its output according to input design criteria. Second, generated outputs contain essential existing information of given site and Seoul. Third, our model visualizes the generated site design in 3D, which provides intuitive images for initial design process. Therefore, the automated urban site design model we suggested is appropriate to examine initial design process visually. We propose that our model reduces time and cost of urban design and site visualization.