Fine-grained traffic information prediction at the turning-level based on low-frequency GNSS trajectory data

The floating-car GNSS trajectory data have been widely used to obtain and predict the urban traffic status in real time, with wide coverage and low deployment cost, and the result has an important supporting role for route decision-making of automatic driving and traffic management. However, the traffic information predicted by the floating-car GNSS data only contains the traffic information on each road segment, ignoring the difference of the traffic flow in different driving directions at the intersection; besides, the accuracy of the traffic information is limited by the GNSS sampling frequency. This paper proposes a turning-level traffic prediction method based on graph convolutional network and low-frequency GNSS trajectory data: first, a queuing-starting-point estimation model is proposed considering vehicle movement pattern; second, a Graph-structure of the turning connection relationship is constructed based on the dual graph theory; finally, to consider spatio-temporal pattern, a traffic prediction model is constructed based on the graph convolutional network. The experimental results show that our method can accurately obtain and predict the traffic speed and queue length at turning-level, and effectively improve the accuracy of traffic prediction by learning the spatio-temporal pattern within the Graph.