Trajectory Prediction for Autonomous Driving System Using Graph Feature Fusion Network

To accurately predict trajectories of surrounding moving objects, we propose a Graph Feature Fusion Network to capture information more fully and effectively on moving objects. Unlike previous works, this work not only designs a 3D graph to incorporate both spatial and temporal edges, but also proposes a multi-level interactive feature fusion network which integrates graph attention model and graph convolution model to obtain the graph feature. Furthermore, encoder-decoder convolutional gated recurrent units are used to predict trajectories of different types of moving objects. ApolloScape dataset is used to evaluate the performances. Results show that our model outperforms several baseline methods on the average displacement error (ADE) and final displacement error (FDE).