Title:
Measuring Street-Level Walkability through Big Image Data and Its Associations with Walking Behavior
Measuring Street-Level Walkability through Big Image Data and Its Associations with Walking Behavior
Author(s)
Koo, Bon Woo
Advisor(s)
Botchwey, Nisha
Guhathakurta, Subhrajit
Guhathakurta, Subhrajit
Editor(s)
Collections
Supplementary to
Permanent Link
Abstract
The built environment characteristics associated with walkability range from neighborhood-level urban form factors to street-level urban design factors. However, many existing walkability measures are primarily based on neighborhood-level factors and lack consideration for street-level factors. Neighborhood-level factors alone can be limited in representing various needs of pedestrians. While pedestrians seek to fulfill their needs for accessibility, safety, comfort, and pleasurability, neighborhood-level factors tend to be limited to capturing the accessibility of the built environment (i.e., having places to go to and being physically connected to those places). The high-order needs (i.e., safety from crime, comfort from vehicular traffic, and aesthetic pleasurability) can be more closely proxied by street-level factors. Also, past studies suggested that certain street-level factors may weaken (or strengthen) the effect of neighborhood-level factors on walking behavior, which can be particularly important for disadvantaged populations who tend to be less responsive to neighborhood-level factors. However, measuring street-level factors often requires extensive manual labor and tends to be resource-intensive, resulting in the omission of street-level factors in widely used walkability measures such as Walk Score.
This dissertation uses street view images and computer vision to overcome these challenges in measuring street-level factors and expands the literature by examining their association with walking mode choice. This dissertation first applies a pre-trained computer vision model to street view images and measure mesoscale (i.e., a midlevel spatial scale between macro and microscale) factors of walkability. It finds that the mesoscale factors have a significant contribution to walking mode choice models, and the contribution is greater than that from neighborhood-level factors. Next, the dissertation develops a method for automatically auditing walkability factors in microscale (i.e., the smallest spatial scale that pertains to the most fine-grain design details and their qualities) using the combination of computer vision, street view images, and geographic information systems. The validation results demonstrate moderate to high reliability between audit results by automated audit method and a trained human auditor. Finally, the dissertation uses automatically audited microscale factors to unpack the reasons for the weaker relationship between neighborhood-level factors and disadvantaged populations’ walking behavior. The result shows that microscale factors play a sizable role in moderating the effect of neighborhood-level factors.
Collectively, this dissertation demonstrates the potential of using street view images and computer vision for research on the built environment-walking relationship and for collecting data on street-level factors over expansive geographic areas, a task that has traditionally been prohibitively expensive. The theoretical and methodological contributions of this dissertation help urban planners and designers understand the physical condition of their cities at street-level and make targeted interventions that are effective and equitable.
Sponsor
Date Issued
2021-07-26
Extent
Resource Type
Text
Resource Subtype
Dissertation