On-Road Directional Trajectory Prediction by Junction-Based Pattern Mining from GPS Data

On-road trajectory prediction has its important applications such as road traffic security and urban road planning. In the past, mathematical models have been formulated for predicting the trajectory of a particular moving object by tracking its latest GPS records. The methods are capable in pinpointing the predicted location in terms of GPS coordinates in the near future. However, in reality, cars and pedestrians do neither move in line-of-sight nor perfect projectile in urban roads. Rather they navigate from junction-to-junction and around building blocks on the roads. In this paper, the authors propose a new computational framework that predicts the next moving direction of on-road trajectory at a junction, based on the probabilities of junction-turns from the aggregated historical traffic patterns. The prediction is constrained by the road formation, the trajectory is tracked by the route that a moving object has travelled, in some abstract format of node pattern. Simple pattern mining is used to match the travelled route with the most frequent routes recorded in the database, for inferring what the next most probable turn will be from the current junction. A simulation experiment is conducted by using Microsoft Trajectory Dataset, that validates the model is efficient and effective.