Virtual Sensors Augment Traffic Data for Urban Management

Urban traffic management systems often face challenges due to the limited spatial coverage of physical sensor networks, constrained by deployment costs and privacy concerns. This sparsity means that machine learning models trained on such data struggle to generalize to unmonitored locations and require retraining when sensor infrastructure changes. A new methodology has been introduced that uses simulations to augment traffic count datasets. This approach replaces physical sensors with "virtual sensors" strategically placed at surrogate locations within the road network. A graph-search heuristic is employed to select these virtual sensor locations, optimizing for both vehicle-flow continuity and traffic-metric similarity between the original and surrogate sites, while ensuring sufficient spatial displacement for data diversity. The method was validated in two Belgian cities, Brussels and Namur, demonstrating that the augmented datasets successfully preserve key traffic characteristics like bimodal daily demand profiles and traffic dynamics. This innovation allows for more robust training of traffic management models, extending their applicability beyond physically monitored areas.