Integrating GIS, deep learning, and environmental sensors for multicriteria evaluation of urban street walkability

Abstract

The walkability of urban streets (WoUS) benefits public health and urban livability, but there is still no consensus on the quantitative and comprehensive evaluation of walkability. However, emerging deep-learning, sensor-network, and GIS technologies suggest the possibility of overcoming previous limitations. Therefore, this study explores a new approach for multicriteria evaluation of WoUS, including physical and perceived walkability. After capturing street-view images, traffic-flow data, and environmental-sensor data covering the streets to be analyzed, we automatically calculate physical walkability from four criteria (walkability index, street quality, traffic flow, and physical environment). We first integrate 2-dimensional analysis of walk score, noise, and light using GIS, then perform 3-dimensional composition calculations of greenery, enclosure, and relative walking width using semantic segmentation, vehicle density using multi-object tracking, and indicator weighting distributions using an analytic hierarchy process. Perceived walkability is evaluated by a hierarchy of walking needs model comprising feasibility, accessibility, safety, comfort, and pleasurability. As an empirical study, we conduct an experiment within Anonymous University for physical and perceived walkability evaluations. The results indicate that with perceived WoUS as a benchmark, the automated physical WoUS results reveal the strengths in data integration and processing that are feasible, reliable, and cost-effective. This was an initial attempt at constructing an integrated research methodology for WoUS from various walking-related variables. This framework could support designers as an assistive tool for generating both quantitative and qualitative judgments from various criteria and recommending walkable pedestrians paths to motivate people to walk further, longer, and more often, rather than providing the shortest paths based on a path-finding algorithm.