An Agent-based Approach to Continuously Detect and Update Road Network Changes Using GPS Trajectories

Abstract Up-to-date road maps are critical in both intelligent transportation and urban management systems. The long cycle of road map generation and updating, and on the other hand, the high rate of expanding public transport, causes the road maps generally be behind the latest actual conditions. The previous studies in road map updating are generally static, applied to the whole study area instead of considering merely the changed road segments, and cannot dynamically respond to the rapid changes in the road networks. This study proposes an agent-based system that dynamically explores newly collected trajectory data and updates road geometry and road type labels. A heuristic change detection algorithm is exploited to detect road network changes (newly added or blocked roads). A road geometry extraction algorithm was developed to digitize newly added roads from the trajectories automatically. Road type labels were predicted using the Random Forest algorithm based on Spatiotemporal properties (speed, acceleration, and standard deviation) of the trajectory data. Finally, the road network map was updated using the newly generated geometry and road type label. GeoLife dataset and OpenStreetMap road network were used to evaluate the proposed approach. Besides, a sensitivity analysis was performed to test the robustness of the results of the road extraction algorithm under various parameter settings. The performances of the road-type prediction algorithms were evaluated using Accuracy, Kappa, Recall, and Precision. Experiments demonstrated the feasibility of the proposed approach for detecting additive road changes and dynamically updating road geometries and types without human intervention.