Trajectories Prediction of Surrounding Vehicles at Urban Intersections

Long-term accurate prediction of surrounding vehicle trajectories is one of the key technologies for unmanned vehicles passing through real urban intersections safely and efficiently. Aiming at the long-time accurate prediction of vehicle trajectories at urban intersections, with the subgrade and real vehicle data acquisition platform, the motion patterns recognition model of target vehicles is established based on the Gaussian mixture model (GMM). The Gaussian process regression (GPR) algorithm is then used to establish the trajectories prediction model for each model extracted from GMM. Finally, the algorithm validation is performed using the subgrade dataset and the real vehicle dataset. The results show that: (1) Gaussian mixture model can effectively extract the motion patterns of vehicles; and (2) Gaussian process regression algorithm is superior to traditional prediction algorithm in long-term trajectories prediction. The findings of the study can provide effective and reliable data support for unmanned vehicles safely passing through intersections.