Automatic Detection of Podotactile Pavements in Urban Environments Through a Deep Learning-Based Approach on MLS/HMLS Point Clouds

Pedestrian accessibility is a critical dimension of sustainable and inclusive transportation systems, yet many cities lack reliable data on infrastructure features that support visually impaired users. Among these, podotactile paving plays a vital role in guiding movement and ensuring safety at intersections and transit nodes. However, tactile paving networks remain largely absent from digital transport inventories and automated mapping pipelines, limiting the ability of cities to systematically assess accessibility conditions. This paper presents a scalable approach for identifying and mapping podotactile areas from mobile and handheld laser scanning data, broadening the scope of data-driven urban modelling to include infrastructure elements critical for visually impaired pedestrians. The framework is evaluated across multiple sensing modalities and geographic contexts, demonstrating robust generalization to diverse transport environments. Across four dataset configurations from Madrid and Mantova, the proposed DeepLabV3+ model achieved podotactile F1-scores ranging from 0.83 to 0.91, with corresponding IoUs between 0.71 and 0.83. The combined Madrid–Mantova dataset reached an F1-score of 0.86 and an IoU of 0.75, highlighting strong cross-city generalization. By addressing a long-standing gap in transportation accessibility research, this study demonstrates that podotactile paving can be systematically extracted and integrated into transport datasets. The proposed approach supports scalable accessibility auditing, enhances digital transport models, and provides planners with actionable data to advance inclusive and equitable mobility.