Neural networks for local monitoring of traffic magnetic sensors

A real-time traffic incident detection algorithm is proposed and applied to the monitoring of a complex road junction in the city of Nancy in France. This algorithm has the potential to provide local monitoring of traffic sensors. Our approach is based on macroscopic traffic flow models, and more precisely on the flow-density relationship. Once this relation is extracted from real traffic data, an admissible region is defined in the flow-density space. Then, the classification properties of neural networks are used to design the monitoring network, which detects and isolates the incidents that disturb the traffic, when the measured data are out of the admissible region. A hierarchical scheme to deal with incidents in large-scale networks is developed as well.