Mohamed R. Ibrahim, James Haworth, Tao Cheng | Cities
Modelling urban systems has interested planners and modellers for decades. Different models have been achieved relying on mathematics, cellular automation, complexity, and scaling. While most of these models tend to be a simplification of reality, today within the paradigm shifts of artificial intelligence across the different fields of science, the applications of computer vision show promising potential in understanding the realistic dynamics of cities. While cities are complex by nature, computer vision shows progress in tackling a variety of complex physical and non-physical visual tasks. In this article, we review the tasks and algorithms of computer vision and their applications in understanding cities. We attempt to subdivide computer vision algorithms into tasks, and cities into layers to show evidence of where computer vision is intensively applied and where further research is needed. We focus on highlighting the potential role of computer vision in understanding urban systems related to the built environment, natural environment, human interaction, transportation, and infrastructure. After showing the diversity of computer vision algorithms and applications, the challenges that remain in understanding the integration between these different layers of cities and their interactions with one another relying on deep learning and computer vision. We also show recommendations for practice and policy-making towards reaching AI-generated urban policies.