Extracting Realistic Pedestrian, Cycling, and Driving Street Networks from OpenStreetMap

Abstract. This paper presents a methodology for extracting realistic and usable pedestrian, cycling, and driving street networks derived from OpenStreetMap (OSM) data. While OSM is a valuable source of information, it is not exempt from inconsistencies and errors, particularly affecting pedestrian and cycling networks, which can lead to inaccurate navigation or unsafe routes. To address this, we propose a methodology which employs a set of filters and post-processing algorithms in a selective removal process to mitigate such inconsistencies and generate networks that are realistic and usable for various applications like routing and analysis. The methodology is designed to be general and globally applicable at the city level. Particular attention is paid to pedestrian networks, addressing the inconsistency where separately mapped sidewalks are not explicitly specified in their corresponding driving streets. An algorithm was designed to detect these cases based on the geometry of the street segments. The methodology also offers an assessment functionality to identify potential inconsistencies in OSM extracted networks, providing support for mapping campaigns seeking to improve its quality. The methodology was applied to five capital cities showing its effectiveness in refining the networks, with larger changes observed in cities with higher mapping quality. Code and examples are available in a GitHub repository.