Assessing urban design: measure unmeasurable things with computer vision

Abstract

Modern cities are now continuously expanding, blurring the concept of the city boundary, accelerating the change of cities' appearance significantly. The goal of urban design is to improve the physical appearance of urban areas to satisfy local citizens and visitors has been challenging recently due to the size and complexness of modern cities in the context of rapid urbanization. This study proposes a computer vision-based approach to assess several factors of urban design quality. We evaluated the enclosure and appearance of natural elements by using deep learning algorithms and street-view images. A resident survey to get the rating score of study locations was also conducted, which be used for the proposed method validation. This method is expected to become a quantitative measure of urban design quality, an abstract and unmeasurable concept.