Improving Methods for Assessing Walkability Using AI-Based Street Level Audits
Walkability matters for public health, equity, and California's climate goals, but the tools used to measure it are surprisingly limited. Popular indices like Walk Score focus on how close homes are to shops and how well streets connect, while ignoring the actual conditions people walk in — whether sidewalks exist, are wide enough, are in good repair, meet accessibility standards, or feel safe and comfortable. This gap means current walkability scores can look good on paper for neighborhoods that, in reality, have missing sidewalks, unsafe crossings, or barriers for people with disabilities — problems that are most common in under-resourced communities.
This project develops a new way to measure walkability using artificial intelligence. The research team will use computer vision to automatically scan street-level photos (from sources like Google Street View) and detect real infrastructure features — sidewalk width and continuity, curb ramps, crosswalks, lighting, shade, and obstacles. An AI language model will then interpret these features, generating both descriptive assessments and numerical scores for comfort, safety, accessibility, and connectivity. These scores will be combined into a new Walkability Quality Index, informed by established pedestrian safety frameworks.
The team will test and validate this tool in San Diego and San Luis Obispo, comparing AI results against in-person walk audits and safety data, and will use explainable-AI techniques to clarify which infrastructure factors matter most in different community types. The result will be a practical, scalable tool that helps planners and policymakers identify where pedestrian infrastructure improvements are most needed, supporting more equitable investment across California communities.
San Diego State University
This project will produce several concrete research and technical outputs. First, a novel "Walkability Quality Index (WQI)" methodology that integrates computer vision, multimodal LLM interpretation, and Pedestrian Level of Traffic Stress (PLTS) and Transit Access and Safety Scoring (TASS) frameworks — a reusable process that other researchers and agencies can apply beyond this project's study areas.
Second, trained "computer vision models" (object detection and semantic segmentation) capable of automatically identifying pedestrian infrastructure features — sidewalks, curb ramps, crosswalks, lighting, shade, and obstructions — from street-level imagery.
Third, structured "datasets" of street-segment-level pedestrian infrastructure features and corresponding WQI scores for roadway segments across San Diego and San Luis Obispo, spanning multiple place typologies (urban, suburban, rural, special-use).
Fourth, a set of "calibrated prompt templates and scoring rubrics" for using multimodal LLMs to assess walkability, tested for consistency across models.
Fifth, an "explainable AI (SHAP-based) analysis framework" identifying which infrastructure characteristics most influence walkability by place type, supporting equity-focused planning decisions.
Finally, the project will produce a "geo-referenced GIS toolkit", an MTI research report, a research brief, conference/journal papers, and course modules for SDSU and Cal Poly curricula — extending the work's reach into practice and education.
This research will give Caltrans, CARB, MPOs, and local agencies a scalable, evidence-based tool for identifying where pedestrian infrastructure investments will have the greatest safety, equity, and climate change benefits. By replacing proxy-based walkability metrics with AI-generated assessments grounded in actual sidewalk, curb ramp, crosswalk, and lighting conditions, agencies can better target limited funding toward the highest-need locations — particularly in under-resourced communities where infrastructure gaps are most severe and often invisible to conventional indices. Practically, this means more defensible prioritization for Active Transportation Program (ATP) and SB 1 funding decisions, stronger safety outcomes through earlier identification of hazardous pedestrian conditions (missing crossings, inadequate lighting, ADA non-compliance), and reduced costs associated with manual walk audits, since AI-based screening can flag priority corridors before fieldwork is deployed. The explainable-AI component further supports policy transparency, helping agencies justify funding decisions with interpretable, feature-level evidence rather than black-box scores. Longer-term, the Walkability Quality Index (WQI) and toolkit could inform statewide performance measures, support CARB's climate and mode-shift goals by improving walking conditions, and be adapted by other states or MPOs. Findings will be disseminated through MTI reports, peer reviewed publications, trainings, and CARB's project evaluation processes, creating a direct pathway from research to funding decisions.
-
Contact Us
San José State University One Washington Square, San Jose, CA 95192 Phone: 408-924-7560 Email: mineta-institute@sjsu.edu
