An Urban Built Environment Analysis Approach for Street View Images Based on Graph Convolutional Neural Networks

• DOI: 10.3390/app14052108
• Publication Date: 2024-03-04T09:36:21Z
• Authors: Changmin Liu; Yang Wang; Weikang Li; Liufeng Tao; Sheng Hu; Mengqi Hao
• Keywords: Technology; QH301-705.5; T; Physics; QC1-999; traffic black spot recognition; graph convolutional neural networks; Engineering (General). Civil engineering (General); Chemistry; 03 medical and health sciences; street view images; 0302 clinical medicine; TA1-2040; Biology (General); QD1-999; urban built environment

Traditionally, research in the field of traffic safety has predominantly focused on two key areas—the identification black spots and analysis accident causation. However, such heavily relies historical records obtained from management department, which often suffer missing or incomplete information. Moreover, these typically offer limited insight into various attributes associated with accidents, thereby posing challenges to comprehensive analyses. Furthermore, collection data incur substantial costs. Consequently, there is a pressing need explore how features urban built environment can effectively facilitate accurate spots, enabling formulation effective strategies support development. In this study, we Kowloon Peninsula Hong Kong, specific focus road intersections as fundamental unit our analysis. We propose leveraging street view images valuable source data, us depict comprehensively. Through utilization models random forest approaches, conduct spot identification, attaining an impressive accuracy rate 87%. To account for impact surrounding adjacent outcomes, adopt node-based approach, treating nodes establishing spatial relationships between them edges. The characterizing at serve node attributes, facilitating construction graph structure representation. By employing graph-based convolutional neural network, enhance methodology, resulting improved 90%. based distinctive environment, analyze underlying causes spots. Our findings highlight significant influence buildings, sky conditions, green spaces, billboards formation Remarkably, observe clear negative correlation while human presence exhibit distinct positive correlation.