Vehicle Trajectory Reconstruction from Global Position System Data with Long Short-term Memory Neural Network

Vehicle trajectory contains massive spatial-temporal information of traffic flow, which is of great significance for the comprehensive deconstruction of urban traffic network. Vehicles equipped with global position system (GPS) detector can collect large-scale, all-weather vehicle trajectory data, which plays an important role in construction of intelligent transportation system (ITS). However, due to the limitation of actual technical level and complex environment, GPS equipment cannot receive valid signals frequently. Therefore, it is necessary to complete missing data so that it can better serve for the intelligent traffic management applications. This article proposes a long shortterm memory neural network (LSTM NN) model to reconstruct vehicle trajectory from GPS data, using LSTM NN to learn the operating pattern of vehicle GPS sequence data. The accuracy of the model is verified on the SUMO microscopic traffic simulation platform. The simulation network is constructed based on the real road network in Zhangjiagang, China, and the signal intersection timing is calibrated according to actual signal timing plan. The results show that the method can achieve satisfactory application effects.

I. Introduction

Nowadays, a great deal of traffic data collected from various types of traffic sensors brings massive benefits to intelligent transportation system (ITS). Among them, as the most comprehensive and specific type of traffic data, vehicle trajectory data contains a lot of temporal and spatial information of traffic flow. It is of great significance to deconstruct the operation of the traffic network based on vehicle trajectory data such as global position system (GPS) data, radio frequency identification (RFID) and license plate recognition (LPR) data. In which GPS data was applied to engineering practice earlier, and was widely studied by many researchers and engineers due to its high acquisition density.