Enhancing Urban Walkability Assessment with Multimodal Large Language Models

Recent advances in computer vision and Convolutional Neural Networks (CNNs) have facilitated the use of street view imagery (SVI) for the automatic assessment of physical and perceived attributes of walkable areas. However, these methods still overlook the broader urban context and fail to capture and communicate to the user the qualitative factors influencing the assessed walkability score. This paper addresses these challenges by leveraging a Multimodal Large Language Model (MLLM) to provide a holistic assessment of walkability, consisting of both quantitative scores and linguistic qualitative insights. This approach offers a more comprehensive understanding of the factors contributing to the walkability score attributed to the image and enhances the interpretability and practical applicability of the assessments for urban planners and policymakers. Preliminary experiments demonstrate that a MLLM-based methodology can effectively capture a diverse range of factors of walkability, suggesting a promising direction for future developments of evaluation tools aimed at supporting the design of pedestrian-friendly environments.