Deep-learning Based Urban Vehicle Trajectory Prediction

Abstract

Recently, as a variety of position sensors are developed, a large amount of urban position data is collected in the urban traffic networks. Based on the data collected through such location sensors, high-resolution urban mobility data of individual users using urban road networks is generated and collected in the transportation systems. Urban mobility data generated by these sensors provide a novel spatio-temporal insights into the mobility patterns of traffic network users and can be used to develop models and strategies to predict traffic flows in urban areas and improve traffic efficiency. This study proposes an algorithm for predicting urban mobility patterns. Deep learning based algorithm is used to train mobility patterns in urban areas and predict mobility. The proposed algorithm is trained and tested using Bluetooth data collected in Brisbane for one year. As a result of evaluating the performance of the algorithm with the test dataset, the proposed algorithm shows an average prediction accuracy of 70% or more.