A Spatial Analysis of Urban Streets under Deep Learning Based on Street View Imagery: Quantifying Perceptual and Elemental Perceptual Relationships

Abstract

Measuring the human perception of urban street space and exploring the street space elements that influence this perception have always interested geographic information and urban planning fields. However, most traditional efforts to investigate urban street perception are based on manual, usually time-consuming, inefficient, and subjective judgments. This shortcoming has a crucial impact on large-scale street spatial analyses. Fortunately, in recent years, deep learning models have gained robust element extraction capabilities for images and achieved very competitive results in semantic segmentation. In this paper, we propose a Street View imagery (SVI)-driven deep learning approach to automatically measure six perceptions of large-scale urban areas, including “safety”, “lively”, “beautiful”, “wealthy”, “depressing”, and “boring”. The model was trained on millions of people’s ratings of SVIs with a high accuracy. First, this paper maps the distribution of the six human perceptions of urban street spaces within the third ring road of Wuhan (appearing as Wuhan later). Secondly, we constructed a multiple linear regression model of “street constituents–human perception” by segmenting the common urban constituents from the SVIs. Finally, we analyzed various objects positively or negatively correlated with the six perceptual indicators based on the multiple linear regression model. The experiments elucidated the subtle weighting relationships between elements in different street spaces and the perceptual dimensions they affect, helping to identify the visual factors that may cause perceptions of an area to be involved. The findings suggested that motorized vehicles such as “cars” and “trucks” can negatively affect people’s perceptions of “safety”, which is different from previous studies. We also examined the influence of the relationships between perceptions, such as “safety” and “wealthy”. Finally, we discussed the “perceptual bias” issue in cities. The findings enhance the understanding of researchers and city managers of the psychological and cognitive processes behind human–street interactions.

Investigating the Impact of Streetscape and Land Surface Temperature on Cycling Behavior

“…This approach allowed for a nuanced understanding of the streetscape, delineating different elements within the urban environment through sophisticated feature extraction and segmentation. The use of the PSPNet model in this study ensured the use of a robust and efficient methodology for analyzing and categorizing streetscape features at the granular level [36]. This enabled our study to provide detailed analysis and fine-grained categorization of street view images, providing reliable data support for the analysis.…”

Predicting Neighborhood-Level Residential Carbon Emissions from Street View Images Using Computer Vision and Machine Learning

“…Semantic segmentation refers to dividing and parsing images into several areas linked with semantic categories [132]. PSPNet has become a commonly used approach in emerging urban studies to extract street canyon characteristics [133][134][135] and has shown state-of-the-art performance on the ADE20K database, achieving an accuracy of over 80% [133,136].…”

Predicting Neighborhood-Level Residential Carbon Emission from Street-view Images Using Computer Vision and Machine Learning

“…Semantic segmentation refers to dividing and parsing images into several areas linked with semantic categories (Guo et al, 2018). The PSPNet has become a commonly used approach by emerging urban studies to extract street canyon characteristics (Zhao et al, 2017;Yuan, Wang and Xu, 2022;Sun et al, 2023) and shown state-of-the-art performance on the ADE20K database, achieving an accuracy of over 80% (Zhao et al, 2017;Zhou et al, 2019).…”