Lane-Level Road Network Modeling Based on the Integration of Satellite Remote Sensing and OpenStreetMap Data

Abstract

Satellite remote sensing provides wide-area urban observations; however, the reliable extraction of lane-level geometry remains difficult because narrow lane markings are affected by spatial resolution, vehicles, buildings, and greenbelts. Although OpenStreetMap (OSM) provides global road vector data, geometric offsets and missing lane attributes limit its direct application to lane-level modeling. Existing approaches that integrate remote sensing images and OSM data for road extraction often treat vehicles as obstructions, overlooking the informative correlation between vehicle distributions and lane topology. To address this, this study utilizes vehicle locations extracted from remote sensing imagery as reference indicators of lane characteristics, and by integrating multitemporal satellite images with OSM road vector data, we propose a Bayesian parameter estimation framework for lane centers, incorporating Markov Chain Monte Carlo (MCMC). Experiments using eight temporal acquisitions over Changchun’s Jingyue District yielded completeness, correctness, and quality values of 91.66%, 91.96%, and 84.86%, respectively, for road centerline extraction, demonstrating its capability for lane-level road network construction.