Computing turn delay in city road network with GPS collected trajectories

In this paper, we aim to mine turn delay at different times and turn types in city road network based on personal GPS collected trajectories. We provide a method to effectively solve the problem for computing turn delay. By using this method, we can rapidly process massive trajectory data, to explore and predict turn delay in city road network. Through map-matching and pre-processing work for trajectory data, we firstly extract turn delay records from the time that people pass across a road intersection. Limited by the range of trajectory collection, these turn delay records cannot cover all road intersection and all different times. Therefore, we secondly propose a prediction model based on Neural Networks to handle these records. In this prediction model we have considered both geography neighborhood effect and topological relationship of road intersections. Finally, we tested the efficiency of this method through cross-validation by using 8986 trajectories derived from 165 pedestrians in a time period of three years. It demonstrates that the proposed method can obtain a higher accuracy of turn delay prediction than traditional methods which usually ignore topological characteristics of road intersections.