Evaluating Pedestrian Environment Using DeepLab Models Based on Street Walkability in Small and Medium-Sized Cities: Case Study in Gaoping, China

Abstract

In small and medium-sized cities of China, walking plays an important role as a green and healthy way to travel. However, the intensification of motorized travel and poor planning of pedestrian transportation systems have resulted in poor travel experiences for residents. To encourage residents to change their mode of travel from motorized transport to greener modes, it is necessary to consider the characteristics of walking travel, design good walking street environments, and increase the advantages of walking in the downtown areas of small and medium-sized cities. In this study, a spatial environment model of a pedestrian street was constructed based on the walking score. Visual perception elements, street function elements, and walking scale elements were acquired by semantic segmentation of Baidu street view images obtained with the DeepLab model. Points of interest (POI) were obtained based on surveys, measurements, and the space syntax. Considering walking distances for small and medium-sized cities, the attenuation coefficient of a reasonable facility distance was adopted to modify the walking score. Based on the comprehensive score obtained, walking paths were divided into four categories: functionally preferred, visually preferred, scale preferred, and environmentally balanced. This categorization provides theoretical support for the design of pedestrian street space environments. Taking the pedestrian street in the city center of Gaoping in Shanxi Province, China as an example, the feasibility of the method and model was verified.