Social inequalities in neighborhood visual walkability: Using Street View imagery and deep learning technologies to facilitate healthy city planning

Abstract

It is of great significance both in theory and in practice to propose an efficient approach to approximating visual walkability given urban residents' growing leisure needs. Recent advancements in sensing and computing technologies provide new opportunities in this regard. This paper first proposes a conceptual framework for understanding street visual walkability and then employs deep learning technologies to segment and extract physical features from Baidu Map Street View (BMSV) imagery using the case of Shenzhen City in China. Guided by this framework, four indicators are calculated based on the segmented imagery and further integrated into the visual walkability index (VWI), whose reliability is validated through manual interpretation and a subjective scoring experiment. Our results show that deep learning technologies achieve higher accuracy in segmenting street view imagery than the traditional K-means clustering algorithm and support vector machine algorithm. Moreover, the developed VWI is effective to measure visual walkability, and it presents great heterogeneity across streets within Shenzhen. Spatial regression further identifies that significant social inequalities are associated with neighborhood visual walkability. According to the findings, implications and suggestions on planning the healthy city are proposed. The methodological procedure is reduplicative and can be applied to other unfeasible or challenging cases.