Exploring the Feasibility of Using “Google Street View” to Assess the Accessibility of Community Environments: Developing Definitions and Observational Protocol for Image Recognition and Classification

Recent developments in image recognition and “Big Data” offer a method for monitoring the physical accessibility of community environments over time. The stimulus characteristics of images showing access features in context are extremely complex, however, and defining access features that are recognizable by humans or machines is a challenging task. Carefully defining access features provides a basis for teaching both humans and machines to recognize those features and to rate the degree of accessibility.

We employed a stage-process to identify access features detectable in images presented by GSV. We created definitions of features, developed a protocol for conducting community observations, and used it to conduct assessments of access images of 14 towns and cities in 9 states and the District of Columbia.

Interobserver agreement averaged 84% on features and 93% on people observed. A combined access score across communities averaged 60%; ranging from 32% to 100%. A scaled “minimum access” score averaged .92; ranging from .59 to 1.0. We observed 158 people, 3 of whom used a mobility device and 13 of whom used other wheeled devices.

Equating these ratings to academic grades suggests that several communities fail to achieve standards of accessibility but do achieve minimal levels of access in their civic cores. Google Street View offers a data source for human observers to assess the accessibility of communities. Researchers should explore the feasibility of using supervised and unsupervised learning to establish machine recognition of access features.