A novel walkability index using google street view and deep learning

Abstract

Despite its wide use across disciplines, the conventional walkability index has limited applicability in planning and design practice because it shows poor agreement with pedestrians’ subjective assessment, which is an important measure of pedestrian-friendliness. One of the key reasons is that the index largely ignores micro-level features that are important in describing walking behavior. To address this gap, we developed a new walkability index that encompasses both micro-level and macro-level attributes using Google Street View (GSV) data and computer vision algorithms. The outperformance of the new index is validated through its alignment with pedestrian-rated overall walking environment satisfaction. The findings of this study underscore the limitations of the Walk-score by revealing a significant disparity with walking environment satisfaction. The validation tests confirm that our new index is superior to the Walk-score because it provides comprehensive reflections of satisfaction. Our results suggest the need for a new walkability index that includes streetscapes as well as macro-level features. Overall, our new index provides urban planners and analysts with an improved metric to capture various built environmental features for planning practice.