Assessing and Comparing the Visual Comfort of Streets across Four Chinese Megacities Using AI-Based Image Analysis and the Perceptive Evaluation Method

Abstract

Environmental perception studies have long been constrained by research scales due to the difficulties in obtaining users’ perceptive data and constructing their relation to environmental attributes. With the help of big data from street view images, this study compares the visual comfort of streets across four Chinese megacities with evidently distinct geographical characteristics. A multi-method approach involving traditional comfort measurements, image analysis based on deep learning algorithms and spatial mapping using geographic information systems was used to investigate the visual components of urban streets at the city scale and their influential mechanisms. In general, the four cities ranked by visual comfort were Beijing first, then Shenzhen, Shanghai and Guangzhou. The results also suggested that the spatial distribution of the four cities’ street visual comfort is obviously different. In Shanghai and Beijing, streets with a higher comfort level are mostly concentrated within the central city, while the highly comfort streets are mostly distributed along the coast and rivers in Shenzhen and Guangzhou. Thus, it is reasonable to speculate that the streets’ visual comfort relates significantly to their urban planning and construction process. Moreover, seven indicators have been identified as influential to street comfort, among which ‘vegetation’, ‘terrain’ and ‘rider’ are positive indicators, while ‘architecture’, ‘pedestrians’, ‘motorcycles’ and ‘bicycles’ have negative influences. Comparing street comfort indicators of the four case study cities, it was observed that ‘vegetation’ and ‘terrain’ have the most consistent positive influences across cities, while the high visibility of ‘building’ on streets is most likely to lead to a low level of perceived comfort. The research outcomes provide applicable cues for large-scale street evaluation research and illustrate an efficient street design approach that can both respond to local characteristics and human perceptive needs.