Subhrajit Guhathakurta / Georgia Institute of Technology
BonWoo Koo / Georgia Institute of Technology
Nisha Botchwey / Georgia Institute of Technology
Many of the existing walkability indices do not include street-level streetscape characteristics in their calculation due to the lack of a scalable way to measure them. This paper demonstrates a new technique based on computer vision to automatically extract information from Google Street View (GSV) images. The information is used to develop indices for representing walkable streetscapes. The GSV indices are used in models of actual walking behavior in Atlanta derived from the 2017 National Household Travel Survey. The ability to predict walking behavior with the GSV indices is compared with Walk Score using logistic regression models. The regression results show that the GSV indices produce better model fits than Walk Score in almost all cases.
With the recent advances in machine learning, particularly computer vision techniques, automatically extracting quantifiable data from Google Street View (GSV) images has become cheaper, scalable, and accurate. It is now possible to detect various objects in images and label each pixel with what that pixel is likely to represent. This study attempts to use one such computer vision technique to analyze GSV images, extract useful information from them, and use it to represent the streetscape characteristics.
The primary objectives of this study are to test the explanatory power of streetscape measurements extracted from GSV images in explaining walking mode choice and to show how it compares with an existing walkability index. In the next section, we present a review of related research on walkability indices and audits, streetscape components relevant to walking behavior, recent advances in the use of Google Street View, and how these discussions relate to the active living literature. The third section describes the data collection and analysis methods. After presenting the indices we developed and the analysis results in the fourth section, the study concludes with a discussion of future research in incorporating streetscapes for walkability measures.