Data-Driven Road User Prediction at Intersections with Connected Sensors

Authors

Julian Bock, MSc., Univ.-Prof. Dr.-Ing. Lutz Eckstein,
Institute for Automotive Engineering (ika), RWTH Aachen University,
Aachen;
Dipl.-Ing. Jens Kotte, Dr.-Ing. Adrian Zlocki,
Forschungsgesellschaft Kraftfahrwesen mbH, Aachen

Summary

The road safety of intersections can be improved through connected infrastructure and vehicle sensors providing information for automated vehicle guidance, intelligent traffic management and adaptive street lighting. Within the German research project I2EASE, systems for improvement of road safety as well as energy efficiency at an intelligent intersection are developed. A key component is the trajectory prediction of road user movement, for which we propose a data-driven approach using Recurrent Neural Networks. The system can predict any type of road users, which is perceived by the connected sensors. Furthermore, the system can learn intersection specific movement patterns. Finally, the prediction accuracy can be estimated permanently and the models can be updated based on new data if the movement patterns have
changed.

Members Login