STGCN For Modeling Vehicle Trajectory in Highway Scenario

This paper proposed a method based on STGCN (Spatial-Temporal Graph Convolutional Network) for predicting vehicles trajectories on highway. This method takes interaction between vehicles and lane information into consideration. In this work, TCN (Temporal Convolutional Network) is used for extracting spatial features, which performs better in long term forecasting than RNN. And this model reduces the amount of input from the perception module, it only requires lane information and vehicles' real-time coordinates, which means it has a low-level demand for computing power.