Clarity or confusion: A review of computer vision street attributes in urban studies and planning

The acceleration of urban imagery data analysis, driven by computer vision (CV), has created noteworthy opportunities for urban studies and planning. Data on street environments with high granularity derived from geo-tagged street views allow urban researchers to obtain geospatial data on greenery, pavement materials, and dimensions, building facades, urban furniture, lighting, vehicle presence, etc. However, how such attributes have been classified and used to address urban studies, planning, or mobility questions remains relatively poorly understood among non-technical researchers. Targeting urban planning and design researchers who do not have a background in CV, this paper reviews planning-relevant attributes that CV approaches of urban streetscapes have delivered to date and examines some of their research applications. We present a systematic analysis of 146 papers scrutinizing 104 street attributes in four groups. By exploring a subcollection of 24 papers, we discuss the effectiveness of those attributes being incorporated into current quantitative urban studies. This study's primary contribution lies in providing a comprehensive summary of CV-driven street attributes, their applications, and the algorithms used, serving as a valuable resource for future urban research. Additionally, we identify key challenges in this field, such as unclear definitions of attributes, a disproportionate emphasis on selecting models and features, and the absence of standardized measurement and definition methods. Furthermore, we offer recommendations for future research directions in this area.