Evolution of historical urban landscape with computer vision and machine learning: A case study of Berlin

Abstract

Previous studies of historical urban landscape usually focused on the qualitative analysis of city planning and development with the social, economic, and political context, but overlooked quantitative analysis from human perceptions. In this study, we objectively measured the seemingly subjective qualities of the eye-level urban landscape perceptions with computer vision and machine learning technologies. We then explored correlations between the measured perceptual qualities and characteristics from the historical context. We chose Berlin as a research object because it is a capital city that owns a rich history of division and reunification, as well as a variety of residential zones built at different periods. We extracted 30 street features from the 150,000 Google Street View Imagery (SVI) dataset with a computer vision algorithm and evaluated eight perceptual qualities including typology, order, ecology, enclosure, aesthetics, richness, accessibility and scale with a machine learning method. Then we defined seven residential districts in Berlin, according to their spatial distributions and construction periods. Through a systematic comparison of perceptual qualities of the seven residential districts, we find perceptual qualities of ecology and enclosure have been improved a lot over a hundred years in Berlin. The housing policies and design codes in different social, economic and political contexts evolved to end overcrowding living conditions, create more open spaces, and develop a better ecological environment. This study enriches our understanding and application of subjective measures of human-centred built environment perception to the evolution of the historical urban landscape. The proposed quantitative framework of subjective human perception measures provides great testimony to evaluate the effects of the implementation of housing and other urban planning policies. Such an automated, multisource, high-throughput and scalable framework can be applied to other cities to determine the personality of the city and assess the impact of urban design and planning policies in streetscape improvement.